Table of Contents

**Learning: Supervised, Unsupervised, Reinforcement Learning**

**What is Machine Learning?**

Machine learning is the study of computer algorithms that improve automatically through experience. It is seen as a subset of artificial intelligence. Machine Learning explores the study and construction of algorithms that can learn from and make predictions on data. You select a model to train and then manually perform feature extraction. Used to devise complex models and algorithms that lend themselves to a prediction which in commercial use is known as predictive analytics.

**Below are the most common Machine Learning use cases and capabilities:**

**What is Supervised Learning? **

Supervised learning is the machine learning task of** inferring a function from labeled training data.** The training data consist of a set of training examples.

**Algorithms:** Support Vector Machines, Regression, Naive Bayes, Decision Trees, K-nearest Neighbor Algorithm and Neural Networks

**Example:** If you built a fruit classifier, the labels will be “this is an orange, this is an apple and this is a banana”, based on showing the classifier examples of apples, oranges and bananas.

**What is Unsupervised learning?**

Unsupervised learning is a type of machine learning algorithm used to **draw inferences from datasets consisting of input data without labelled responses.**

**Algorithms:** Clustering, Anomaly Detection, Neural Networks and Latent Variable Models

**Example:** In the same example, a fruit clustering will categorize as “fruits with soft skin and lots of dimples”, “fruits with shiny hard skin” and “elongated yellow fruits”.

**Explain the difference between supervised and unsupervised machine learning?**

In supervised machine learning algorithms, we have to provide labeled data, for example, prediction of stock market prices, whereas in unsupervised we need not have labeled data, for example, classification of emails into spam and non-spam.

**What is deep learning, and how does it contrast with other machine learning algorithms?**

Deep learning is a subset of machine learning that is concerned with neural networks: how to use backpropagation and certain principles from neuroscience to more accurately model large sets of unlabelled or semi-structured data. In that sense, deep learning represents an unsupervised learning algorithm that learns representations of data through the use of neural nets.

**What are the different Algorithm techniques in Machine Learning?**

The different types of techniques in Machine Learning are

● Supervised Learning

● Unsupervised Learning

● Semi-supervised Learning

● Reinforcement Learning

● Transduction

● Learning to Learn

**What’s the difference between a generative and discriminative model?**

A generative model will learn categories of data while a discriminative model will simply learn the distinction between different categories of data. Discriminative models will generally outperform generative models on classification tasks.

**What Are the Applications of Supervised Machine Learning in Modern Businesses?**

Applications of supervised machine learning include:

● **Email Spam Detection**

Here we train the model using historical data that consists of emails categorized as spam or not spam. This labeled information is fed as input to the model.

● **Healthcare Diagnosis**

By providing images regarding a disease, a model can be trained to detect if a person is suffering from the disease or not.

● **Sentiment Analysis**

This refers to the process of using algorithms to mine documents and determine whether they’re positive, neutral, or negative in sentiment.

● **Fraud Detection**

Training the model to identify suspicious patterns, we can detect instances of possible fraud.

**What Is Semi-supervised Machine Learning?**

Supervised learning uses data that is completely labeled, whereas unsupervised learning uses no training data.

In the case of semi-supervised learning, the training data contains a small amount of labeled data and a large amount of unlabeled data.

**What Are Unsupervised Machine Learning Techniques?**

There are two techniques used in unsupervised learning: clustering and association.

**Clustering**

● Clustering problems involve data to be divided into subsets. These subsets, also called clusters, contain data that are similar to each other. Different clusters reveal different details about the objects, unlike classification or regression.

**Association**

● In an association problem, we identify patterns of associations between different variables or items.

● For example, an eCommerce website can suggest other items for you to buy, based on the prior purchases that you have made, spending habits, items in your wish list, other customers’ purchase habits, and so on.

**What evaluation approaches would you work to gauge the effectiveness of a machine learning model?**

You would first split the dataset into training and test sets, or perhaps use cross-validation techniques to further segment the dataset into composite sets of training and test sets within the data. You should then implement a choice selection of performance metrics: here is a fairly comprehensive list. You could use measures such as the F1 score, the accuracy, and the confusion matrix. What’s important here is to demonstrate that you understand the nuances of how a model is measured and how to choose the right performance measures for the right situations.

**What Are the Three Stages of Building a Model in Machine Learning?**

The three stages of building a machine learning model are:

● Model Building Choose a suitable algorithm for the model and train it according to the requirement

● Model Testing Check the accuracy of the model through the test data

● Applying the Mode Make the required changes after testing and use the final model for real-time projects. Here, it’s important to remember that once in a while, the model needs to be checked to make sure it’s working correctly. It should be modified to make sure that it is up-to-date.

**What is Deep Learning?**

Deep Learning is nothing but a paradigm of machine learning which has shown incredible promise in recent years. This is because of the fact that Deep Learning shows a great analogy with the functioning of the neurons in the human brain.

**What is the difference between machine learning and deep learning?**

Machine learning is a field of computer science that gives computers the ability to learn without being explicitly programmed. Machine learning can be categorized in the following four categories.

1. Supervised machine learning,

2. Semi-supervised machine learning,

3. Unsupervised machine learning,

4. Reinforcement learning.

Deep Learning is a subfield of machine learning concerned with algorithms inspired by the structure and function of the brain called artificial neural networks.

• The main difference between deep learning and machine learning is due to the way data is

presented in the system. Machine learning algorithms almost always require structured data, while deep learning networks rely on layers of ANN (artificial neural networks).

• Machine learning algorithms are designed to “learn” to act by understanding labeled data and then use it to produce new results with more datasets. However, when the result is incorrect, there is a need to “teach them”. Because machine learning algorithms require bulleted data, they are not suitable for solving complex queries that involve a huge amount of data.

• Deep learning networks do not require human intervention, as multilevel layers in neural

networks place data in a hierarchy of different concepts, which ultimately learn from their own mistakes. However, even they can be wrong if the data quality is not good enough.

• Data decides everything. It is the quality of the data that ultimately determines the quality of the result.

• Both of these subsets of AI are somehow connected to data, which makes it possible to represent a certain form of “intelligence.” However, you should be aware that deep learning requires much more data than a traditional machine learning algorithm. The reason for this is that deep learning networks can identify different elements in neural network layers only when more than a million data points interact. Machine learning algorithms, on the other hand, are capable of learning by pre-programmed criteria.

**Can you explain the differences between supervised, unsupervised, and reinforcement learning?**

In supervised learning, we train a model to learn the relationship between input data and output

data. We need to have labeled data to be able to do supervised learning.

With unsupervised learning, we only have unlabeled data. The model learns a representation of the data. Unsupervised learning is frequently used to initialize the parameters of the model when we have a lot of unlabeled data and a small fraction of labeled data. We first train an unsupervised model and, after that, we use the weights of the model to train a supervised model. In reinforcement learning, the model has some input data and a reward depending on the output of the model. The model learns a policy that maximizes the reward. Reinforcement learning has been applied successfully to strategic games such as Go and even classic Atari video games.

**What is the reason for the popularity of Deep Learning in recent times? **

Now although Deep Learning has been around for many years, the major breakthroughs from these techniques came just in recent years. This is because of two main reasons:

• The increase in the amount of data generated through various sources

• The growth in hardware resources required to run these models

GPUs are multiple times faster and they help us build bigger and deeper deep learning models in comparatively less time than we required previously

**What is reinforcement learning?**

Reinforcement Learning allows to take actions to max cumulative reward. It learns by trial and error through reward/penalty system. Environment rewards agent so by time agent makes better decisions.

Ex: robot=agent, maze=environment. Used for complex tasks (self-driving cars, game AI).

RL is a series of time steps in a Markov Decision Process:

1. Environment: space in which RL operates

2. State: data related to past action RL took

3. Action: action taken

4. Reward: number taken by agent after last action

5. Observation: data related to environment: can be visible or partially shadowed

**Explain Ensemble learning.**

In ensemble learning, many base models like classifiers and regressors are generated and combined together so that they give better results. It is used when we build component classifiers that are accurate and independent. There are sequential as well as parallel ensemble methods.

**What are the parametric models? Give an example.**

Parametric models are those with a finite number of parameters. To predict new data, you only need to know the parameters of the model. Examples include linear regression, logistic regression, and linear SVMs.

Non-parametric models are those with an unbounded number of parameters, allowing for more flexibility. To predict new data, you need to know the parameters of the model and the state of the data that has been observed. Examples include decision trees, k-nearest neighbors, and topic models using latent Dirichlet analysis.

**What are support vector machines?**

Support vector machines are supervised learning algorithms used for classification and regression analysis.

**What is batch statistical learning?**

Statistical learning techniques allow learning a function or predictor from a set of observed data that can make predictions about unseen or future data. These techniques provide guarantees on the performance of the learned predictor on the future unseen data based on a statistical assumption on the data generating process.

**What Will Happen If the Learning Rate is Set inaccurately (Too Low or Too High)?**

When your learning rate is too low, training of the model will progress very slowly as we are making minimal updates to the weights. It will take many updates before reaching the minimum point.

If the learning rate is set too high, this causes undesirable divergent behavior to the loss function due to drastic updates in weights. It may fail to converge (model can give a good output) or even diverge (data is too chaotic for the network to train).

**What Is The Difference Between Epoch, Batch, and Iteration in Deep Learning? **

• **Epoch** – Represents one iteration over the entire dataset (everything put into the training model).

• **Batch** – Refers to when we cannot pass the entire dataset into the neural network at once, so we divide the dataset into several batches.

• **Iteration** – if we have 10,000 images as data and a batch size of 200. then an epoch should run 50 iterations (10,000 divided by 50).

**Why Is Tensorflow the Most Preferred Library in Deep Learning?**

Tensorflow provides both C++ and Python APIs, making it easier to work on and has a faster compilation time compared to other Deep Learning libraries like Keras and Torch. Tensorflow supports both CPU and GPU computing devices.

**What Do You Mean by Tensor in Tensorflow?**

A tensor is a mathematical object represented as arrays of higher dimensions. These arrays of data with different dimensions and ranks fed as input to the neural network are called “Tensors.”

**Explain a Computational Graph.**

Everything in TensorFlow is based on creating a computational graph. It has a network of nodes where each node operates, Nodes represent mathematical operations, and edges represent tensors. Since data flows in the form of a graph, it is also called a “DataFlow Graph.”

**Cognition: Reasoning on top of data (Regression, Classification, Pattern Recognition)**

**What is the difference between classification and regression?**

Classification is used to produce discrete results, classification is used to classify data into some specific categories. For example, classifying emails into spam and non-spam categories.

Whereas, We use regression analysis when we are dealing with continuous data, for example predicting stock prices at a certain point in time.

**Explain the Bias-Variance Tradeoff.**

Predictive models have a tradeoff between bias (how well the model fits the data) and variance (how much the model changes based on changes in the inputs).

Simpler models are stable (low variance) but they don’t get close to the truth (high bias).

More complex models are more prone to overfitting (high variance) but they are expressive enough to get close to the truth (low bias). The best model for a given problem usually lies somewhere in the middle.

**What is the difference between stochastic gradient descent (SGD) and gradient descent (GD)?**

Both algorithms are methods for finding a set of parameters that minimize a loss function by evaluating parameters against data and then making adjustments.

In standard gradient descent, you’ll evaluate all training samples for each set of parameters.

This is akin to taking big, slow steps toward the solution.

In stochastic gradient descent, you’ll evaluate only 1 training sample for the set of parameters before updating them. This is akin to taking small, quick steps toward the solution.

**How Can You Choose a Classifier Based on a Training Set Data Size?**

When the training set is small, a model that has a right bias and low variance seems to work better because they are less likely to overfit. For example, Naive Bayes works best when the training set is large. Models with low bias and high variance tend to perform better as they work fine with complex relationships.

**Explain Latent Dirichlet Allocation (LDA)**

Latent Dirichlet Allocation (LDA) is a common method of topic modeling, or classifying documents by subject matter.

LDA is a generative model that represents documents as a mixture of topics that each have their own probability distribution of possible words.

The “Dirichlet” distribution is simply a distribution of distributions. In LDA, documents are distributions of topics that are distributions of words.

**Explain Principle Component Analysis (PCA)**

PCA is a method for transforming features in a dataset by combining them into uncorrelated linear combinations.

These new features, or principal components, sequentially maximize the variance represented (i.e. the first principal component has the most variance, the second principal component has the second most, and so on).

As a result, PCA is useful for dimensionality reduction because you can set an arbitrary variance cutoff.

**PCA is a dimensionality reduction technique** that enables you to identify the correlations and patterns in the dataset so that it can be transformed into a dataset of significantly lower dimensions without any loss of important information.

• It is an unsupervised statistical technique used to examine the interrelations among a set of variables. It is also known as a general factor analysis where regression determines a line of best fit.

• It works on a condition that while the data in a higher-dimensional space is mapped to data in a lower dimension space, the variance or spread of the data in the lower dimensional space should be maximum.

**PCA is carried out in the following steps**

1. Standardization of Data

2. Computing the covariance matrix

3. Calculation of the eigenvectors and eigenvalues

4. Computing the Principal components

5. Reducing the dimensions of the Data.

**Reference: Here**

**What’s the F1 score? How would you use it?**

The F1 score is a measure of a model’s performance. It is a weighted average of the precision and recall of a model, with results tending to 1 being the best, and those tending to 0 being the worst. You would use it in classification tests where true negatives don’t matter much.

**When should you use classification over regression?**

Classification produces discrete values and dataset to strict categories, while regression gives you continuous results that allow you to better distinguish differences between individual points.

You would use classification over regression if you wanted your results to reflect the belongingness of data points in your dataset to certain explicit categories (ex: If you wanted to know whether a name was male or female rather than just how correlated they were with male and female names.)

**How do you ensure you’re not overfitting with a model?**

This is a simple restatement of a fundamental problem in machine learning: the possibility of overfitting training data and carrying the noise of that data through to the test set, thereby providing inaccurate generalizations.

There are three main methods to avoid overfitting:

1- Keep the model simpler: reduce variance by taking into account fewer variables and parameters, thereby removing some of the noise in the training data.

2- Use cross-validation techniques such as k-folds cross-validation.

3- Use regularization techniques such as LASSO that penalize certain model parameters if they’re likely to cause overfitting.

**How Will You Know Which Machine Learning Algorithm to Choose for Your Classification Problem?**

While there is no fixed rule to choose an algorithm for a classification problem, you can follow these guidelines:

● If accuracy is a concern, test different algorithms and cross-validate them

● If the training dataset is small, use models that have low variance and high bias

● If the training dataset is large, use models that have high variance and little bias

**Why is Area Under ROC Curve (AUROC) better than raw accuracy as an out-of-sample evaluation metric?**

AUROC is robust to class imbalance, unlike raw accuracy.

For example, if you want to detect a type of cancer that’s prevalent in only 1% of the population, you can build a model that achieves 99% accuracy by simply classifying everyone has cancer-free.

**What are the advantages and disadvantages of neural networks?**

Advantages: Neural networks (specifically deep NNs) have led to performance breakthroughs for unstructured datasets such as images, audio, and video. Their incredible flexibility allows them to learn patterns that no other ML algorithm can learn.

Disadvantages: However, they require a large amount of training data to converge. It’s also difficult to pick the right architecture, and the internal “hidden” layers are incomprehensible.

**Define Precision and Recall.**

Precision

● Precision is the ratio of several events you can correctly recall to the total number of events you recall (mix of correct and wrong recalls).

● Precision = (True Positive) / (True Positive + False Positive)

Recall

● A recall is the ratio of a number of events you can recall the number of total events.

● Recall = (True Positive) / (True Positive + False Negative)

**What Is Decision Tree Classification?**

A decision tree builds classification (or regression) models as a tree structure, with datasets broken up into ever-smaller subsets while developing the decision tree, literally in a tree-like way with branches and nodes. Decision trees can handle both categorical and numerical data.

**What Is Pruning in Decision Trees, and How Is It Done?**

Pruning is a technique in machine learning that reduces the size of decision trees. It reduces the complexity of the final classifier, and hence improves predictive accuracy by the reduction of overfitting.

Pruning can occur in:

● Top-down fashion. It will traverse nodes and trim subtrees starting at the root

● Bottom-up fashion. It will begin at the leaf nodes

There is a popular pruning algorithm called reduced error pruning, in which:

● Starting at the leaves, each node is replaced with its most popular class

● If the prediction accuracy is not affected, the change is kept

● There is an advantage of simplicity and speed

**What Is a Recommendation System?**

Anyone who has used Spotify or shopped at Amazon will recognize a recommendation system:

It’s an information filtering system that predicts what a user might want to hear or see based on choice patterns provided by the user.

**What Is Kernel SVM?**

Kernel SVM is the abbreviated version of the kernel support vector machine. Kernel methods are a class of algorithms for pattern analysis, and the most common one is the kernel SVM.

What Are Some Methods of Reducing Dimensionality?

You can reduce dimensionality by combining features with feature engineering, removing collinear features, or using algorithmic dimensionality reduction.

Now that you have gone through these machine learning interview questions, you must have got an idea of your strengths and weaknesses in this domain.

**How is KNN different from k-means clustering?**

K-Nearest Neighbors is a supervised classification algorithm, while k-means clustering is an unsupervised clustering algorithm. While the mechanisms may seem similar at first, what this really means is that in order for K-Nearest Neighbors to work, you need labeled data you want to classify an unlabeled point into (thus the nearest neighbor part). K-means clustering requires only a set of unlabeled points and a threshold: the algorithm will take unlabeled points and gradually learn how to cluster them into groups by computing the mean of the distance between different points.

**What are difference between Data Mining and Machine learning?**

Machine learning relates to the study, design, and development of the algorithms that give computers the capability to learn without being explicitly programmed. While data mining can be defined as the process in which the unstructured data tries to extract knowledge or unknown interesting patterns. During this processing machine, learning algorithms are used.

**What is “Naive” in a Naive Bayes?**

Reference: Naive Bayes Classifier on Wikipedia

Naive Bayes methods are a set of supervised learning algorithms based on applying Bayes’ theorem with the “naive” assumption of conditional independence between every pair of features given the value of the class variable. Bayes’ theorem states the following relationship, given class variable y and dependent feature vector X1through Xn:

**What is PCA (Principal Component Analysis)? When do you use it?**

Reference: PCA on wikipedia

Principal component analysis (PCA) is a statistical method used in Machine Learning. It consists in projecting data in a higher dimensional space into a lower dimensional space by maximizing the variance of each dimension.

The process works as following. We define a matrix A with > rows (the single observations of a dataset – in a tabular format, each single row) and @ columns, our features. For this matrix we construct a variable space with as many dimensions as there are features. Each feature represents one coordinate axis. For each feature, the length has been standardized according to a scaling criterion, normally by scaling to unit variance. It is determinant to scale the features to a common scale, otherwise the features with a greater magnitude will weigh more in determining the principal components. Once plotted all the observations and computed the mean of each variable, that mean will be represented by a point in the center of our plot (the center of gravity). Then, we subtract each observation with the mean, shifting the coordinate system with the center in the origin. The best fitting line resulting is the line that best accounts for the shape of the point swarm. It represents the maximum variance direction in the data. Each observation may be projected onto this line in order to get a coordinate value along the PC-line. This value is known as a score. The next best-fitting line can be similarly chosen from directions perpendicular to the first.

Repeating this process yields an orthogonal basis in which different individual dimensions of the data are uncorrelated. These basis vectors are called principal components.

PCA is mostly used as a tool in exploratory data analysis and for making predictive models. It is often used to visualize genetic distance and relatedness between populations.

**SVM (Support Vector Machine) algorithm**

Reference: SVM on wikipedia

Classifying data is a common task in machine learning. Suppose some given data points each belong to one of two classes, and the goal is to decide which class a new data point will be in. In the case of supportvector machines, a data point is viewed as a p-dimensional vector (a list of p numbers), and we want to know whether we can separate such points with a (p − 1)-dimensional hyperplane. This is called a linear classifier. There are many hyperplanes that might classify the data. One reasonable choice as the best hyperplane is the one that represents the largest separation, or margin, between the two classes. So, we

choose the hyperplane so that the distance from it to the nearest data point on each side is maximized. If such a hyperplane exists, it is known as the maximum-margin hyperplane and the linear classifier it defines is known as a maximum-margin classifier; or equivalently, the perceptron of optimal stability. The best hyper plane that divides the data is H_{3.}

- SVMs are helpful in text and hypertext categorization, as their application can significantly reduce the need for labeled training instances in both the standard inductive and transductive settings.
- Some methods for shallow semantic parsing are based on support vector machines.
- Classification of images can also be performed using SVMs. Experimental results show that SVMs achieve significantly higher search accuracy than traditional query refinement schemes after just three to four rounds of relevance feedback.
- Classification of satellite data like SAR data using supervised SVM.
- Hand-written characters can be recognized using SVM.

**What are the support vectors in SVM? **

In the diagram, we see that the sketched lines mark the distance from the classifier (the hyper plane) to the closest data points called the support vectors (darkened data points). The distance between the two thin lines is called the margin.

To extend SVM to cases in which the data are not linearly separable, we introduce the hinge loss function, max (0, 1 – y_{i}(w∙ x_{i} − b)). This function is zero if x lies on the correct side of the margin. For data on the wrong side of the margin, the function’s value is proportional to the distance from the margin.

**What are the different kernels in SVM?**

There are four types of kernels in SVM.

1. LinearKernel

2. Polynomial kernel

3. Radial basis kernel

4. Sigmoid kernel

**What are the most known ensemble algorithms?**

Reference: Ensemble Algorithms

The most popular trees are: AdaBoost, Random Forest, and eXtreme Gradient Boosting (XGBoost).

AdaBoost is best used in a dataset with low noise, when computational complexity or timeliness of results is not a main concern and when there are not enough resources for broader hyperparameter tuning due to lack of time and knowledge of the user.

Random forests should not be used when dealing with time series data or any other data where look-ahead bias should be avoided, and the order and continuity of the samples need to be ensured. This algorithm can handle noise relatively well, but more knowledge from the user is required to adequately tune the algorithm compared to AdaBoost.

The main advantages of XGBoost is its lightning speed compared to other algorithms, such as AdaBoost, and its regularization parameter that successfully reduces variance. But even aside from the regularization parameter, this algorithm leverages a learning rate (shrinkage) and subsamples from the features like random forests, which increases its ability to generalize even further. However, XGBoost is more difficult to understand, visualize and to tune compared to AdaBoost and random forests. There is a multitude of hyperparameters that can be tuned to increase performance.

**What are Artificial Neural Networks?**

Artificial Neural networks are a specific set of algorithms that have revolutionized machine learning. They are inspired by biological neural networks. Neural Networks can adapt to changing the input, so the network generates the best possible result without needing to redesign the output criteria.

Artificial Neural Networks works on the same principle as a biological Neural Network. It consists of inputs which get processed with weighted sums and Bias, with the help of Activation Functions.

**How Are Weights Initialized in a Network?**

There are two methods here: we can either initialize the weights to zero or assign them randomly.

Initializing all weights to 0: This makes your model similar to a linear model. All the neurons and every layer perform the same operation, giving the same output and making the deep net useless.

Initializing all weights randomly: Here, the weights are assigned randomly by initializing them very close to 0. It gives better accuracy to the model since every neuron performs different computations. This is the most commonly used method.

**What Is the Cost Function? **

Also referred to as “loss” or “error,” cost function is a measure to evaluate how good your model’s performance is. It’s used to compute the error of the output layer during backpropagation. We push that error backwards through the neural network and use that during the different training functions.

The most known one is the mean sum of squared errors.

**What Are Hyperparameters?**

With neural networks, you’re usually working with hyperparameters once the data is formatted correctly.

A hyperparameter is a parameter whose value is set before the learning process begins. It determines how a network is trained and the structure of the network (such as the number of hidden units, the learning rate, epochs, batches, etc.).

**What Are the Different Layers on CNN?**

Reference: Layers of CNN

The Convolutional neural networks are regularized versions of multilayer perceptron (MLP). They were developed based on the working of the neurons of the animal visual cortex.

**The objective of using the CNN:**

The idea is that you give the computer this array of numbers and it will output numbers that describe the probability of the image being a certain class (.80 for a cat, .15 for a dog, .05 for a bird, etc.). It works similar to how our brain works. When we look at a picture of a dog, we can classify it as such if the picture has identifiable features such as paws or 4 legs. In a similar way, the computer is able to perform image classification by looking for low-level features such as edges and curves and then building up to more abstract concepts through a series of convolutional layers. The computer uses low-level features obtained at the initial levels to generate high-level features such as paws or eyes to identify the object.

**There are four layers in CNN:**

1. Convolutional Layer – the layer that performs a convolutional operation, creating several smaller picture windows to go over the data.

2. Activation Layer (ReLU Layer) – it brings non-linearity to the network and converts all the negative pixels to zero. The output is a rectified feature map. It follows each convolutional layer.

3. Pooling Layer – pooling is a down-sampling operation that reduces the dimensionality of the feature map. Stride = how much you slide, and you get the max of the n x n matrix

4. Fully Connected Layer – this layer recognizes and classifies the objects in the image.

**What Is Pooling on CNN, and How Does It Work?**

Pooling is used to reduce the spatial dimensions of a CNN. It performs down-sampling operations to reduce the dimensionality and creates a pooled feature map by sliding a filter matrix over the input matrix.

**What are Recurrent Neural Networks (RNNs)? **

Reference: RNNs

RNNs are a type of artificial neural networks designed to recognize the pattern from the sequence of data such as Time series, stock market and government agencies etc.

Recurrent Neural Networks (RNNs) add an interesting twist to basic neural networks. A vanilla neural network takes in a fixed size vector as input which limits its usage in situations that involve a ‘series’ type input with no predetermined size.

RNNs are designed to take a series of input with no predetermined limit on size. One could ask what’s\ the big deal, I can call a regular NN repeatedly too?

Sure can, but the ‘series’ part of the input means something. A single input item from the series is related to others and likely has an influence on its neighbors. Otherwise it’s just “many” inputs, not a “series” input (duh!).

Recurrent Neural Network remembers the past and its decisions are influenced by what it has learnt from the past. Note: Basic feed forward networks “remember” things too, but they remember things they learnt during training. For example, an image classifier learns what a “1” looks like during training and then uses that knowledge to classify things in production.

While RNNs learn similarly while training, in addition, they remember things learnt from prior input(s) while generating output(s). RNNs can take one or more input vectors and produce one or more output vectors and the output(s) are influenced not just by weights applied on inputs like a regular NN, but also by a “hidden” state vector representing the context based on prior input(s)/output(s). So, the same input could produce a different output depending on previous inputs in the series.

In summary, in a vanilla neural network, a fixed size input vector is transformed into a fixed size output vector. Such a network becomes “recurrent” when you repeatedly apply the transformations to a series of given input and produce a series of output vectors. There is no pre-set limitation to the size of the vector. And, in addition to generating the output which is a function of the input and hidden state, we update the hidden state itself based on the input and use it in processing the next input.

**What is the role of the Activation Function?**

The Activation function is used to introduce non-linearity into the neural network helping it to learn more complex function. Without which the neural network would be only able to learn linear function which is a linear combination of its input data. An activation function is a function in an artificial neuron that delivers an output based on inputs.

**Machine Learning libraries for various purposes**

**What is an Auto-Encoder?**

Reference: Auto-Encoder

Auto-encoders are simple learning networks that aim to transform inputs into outputs with the minimum possible error. This means that we want the output to be as close to input as possible. We add a couple of layers between the input and the output, and the sizes of these layers are smaller than the input layer. The auto-encoder receives unlabeled input which is then encoded to reconstruct the input.

An autoencoder is a type of artificial neural network used to learn efficient data coding in an unsupervised manner. The aim of an autoencoder is to learn a representation (encoding) for a set of data, typically for dimensionality reduction, by training the network to ignore signal “noise”. Along with the reduction side, a reconstructing side is learnt, where the autoencoder tries to generate from the reduced encoding a representation as close as possible to its original input, hence its name. Several variants exist to the basic model, with the aim of forcing the learned representations of the input to assume useful properties.

Autoencoders are effectively used for solving many applied problems, from face recognition to acquiring the semantic meaning of words.

**What is a Boltzmann Machine?**

Boltzmann machines have a simple learning algorithm that allows them to discover interesting features that represent complex regularities in the training data. The Boltzmann machine is basically used to optimize the weights and the quantity for the given problem. The learning algorithm is very slow in networks with many layers of feature detectors. “Restricted Boltzmann Machines” algorithm has a single layer of feature detectors which makes it faster than the rest.

**What Is Dropout and Batch Normalization?**

Dropout is a technique of dropping out hidden and visible nodes of a network randomly to prevent overfitting of data (typically dropping 20 per cent of the nodes). It doubles the number of iterations needed to converge the network. It used to avoid overfitting, as it increases the capacity of generalization.

Batch normalization is the technique to improve the performance and stability of neural networks by normalizing the inputs in every layer so that they have mean output activation of zero and standard deviation of one

**Why Is TensorFlow the Most Preferred Library in Deep Learning?**

TensorFlow provides both C++ and Python APIs, making it easier to work on and has a faster compilation time compared to other Deep Learning libraries like Keras and PyTorch. TensorFlow supports both CPU and GPU computing devices.

**What is Tensor in TensorFlow?**

A tensor is a mathematical object represented as arrays of higher dimensions. Think of a n-D matrix. These arrays of data with different dimensions and ranks fed as input to the neural network are called “Tensors.”

**What is the Computational Graph?**

Everything in a TensorFlow is based on creating a computational graph. It has a network of nodes where each node operates. Nodes represent mathematical operations, and edges represent tensors. Since data flows in the form of a graph, it is also called a “DataFlow Graph.”

**How is logistic regression done? **

Logistic regression measures the relationship between the dependent variable (our label of what we want to predict) and one or more independent variables (our features) by estimating probability using its underlying logistic function (sigmoid).

**Explain the steps in making a decision tree. **

1. Take the entire data set as input

2. Calculate entropy of the target variable, as well as the predictor attributes

3. Calculate your information gain of all attributes (we gain information on sorting different objects from each other)

4. Choose the attribute with the highest information gain as the root node

5. Repeat the same procedure on every branch until the decision node of each branch is finalized

For example, let’s say you want to build a decision tree to decide whether you should accept or decline a job offer. The decision tree for this case is as shown:

It is clear from the decision tree that an offer is accepted if:

• Salary is greater than $50,000

• The commute is less than an hour

• Coffee is offered

**How do you build a random forest model?**

A random forest is built up of a number of decision trees. If you split the data into different packages and make a decision tree in each of the different groups of data, the random forest brings all those trees together.

Steps to build a random forest model:

1. Randomly select ; features from a total of = features where k<< m

2. Among the ; features, calculate the node D using the best split point

3. Split the node into daughter nodes using the best split

4. Repeat steps two and three until leaf nodes are finalized

5. Build forest by repeating steps one to four for > times to create > number of trees

**Differentiate between univariate, bivariate, and multivariate analysis. **

**Univariate data** contains only one variable. The purpose of the univariate analysis is to describe the data and find patterns that exist within it.

The patterns can be studied by drawing conclusions using mean, median, mode, dispersion or range, minimum, maximum, etc.

**Bivariate data** involves two different variables. The analysis of this type of data deals with causes and relationships and the analysis is done to determine the relationship between the two variables.

Here, the relationship is visible from the table that temperature and sales are directly proportional to each other. The hotter the temperature, the better the sales.

**Multivariate data** involves three or more variables, it is categorized under multivariate. It is similar to a bivariate but contains more than one dependent variable.

**Example: data for house price prediction**

The patterns can be studied by drawing conclusions using mean, median, and mode, dispersion or range, minimum, maximum, etc. You can start describing the data and using it to guess what the price of the house will be.

**What are the feature selection methods used to select the right variables?**

There are two main methods for feature selection.**Filter Methods**

This involves:

• Linear discrimination analysis

• ANOVA

• Chi-Square

The best analogy for selecting features is “bad data in, bad answer out.” When we’re limiting or selecting the features, it’s all about cleaning up the data coming in.

**Wrapper Methods**

This involves:

• Forward Selection: We test one feature at a time and keep adding them until we get a good fit

• Backward Selection: We test all the features and start removing them to see what works

better

• Recursive Feature Elimination: Recursively looks through all the different features and how they pair together

**Wrapper methods are very labor-intensive**, and high-end computers are needed if a lot of data analysis is performed with the wrapper method.

**You are given a data set consisting of variables with more than 30 percent missing values. How will you deal with them? **

If the data set is large, we can just simply remove the rows with missing data values. It is the quickest way; we use the rest of the data to predict the values.

For smaller data sets, we can impute missing values with the mean, median, or average of the rest of the data using pandas data frame in python. There are different ways to do so, such as: df.mean(), df.fillna(mean)

Other option of imputation is using KNN for numeric or classification values (as KNN just uses k closest values to impute the missing value).

Q76: How will you calculate the Euclidean distance in Python?

plot1 = [1,3]

plot2 = [2,5]

The Euclidean distance can be calculated as follows:

**euclidean_distance** = sqrt((plot1[0]-plot2[0])**2 + (plot1[1]- plot2[1])**2)

**What are dimensionality reduction and its benefits? **

Dimensionality reduction refers to the process of converting a data set with vast dimensions into data with fewer dimensions (fields) to convey similar information concisely.

This reduction helps in compressing data and reducing storage space. It also reduces computation time as fewer dimensions lead to less computing. It removes redundant features; for example, there’s no point in storing a value in two different units (meters and inches).

**How should you maintain a deployed model?**

The steps to maintain a deployed model are (CREM):

**1. Monitor:** constant monitoring of all models is needed to determine their performance accuracy.

When you change something, you want to figure out how your changes are going to affect things.

This needs to be monitored to ensure it’s doing what it’s supposed to do.**2. Evaluate:** evaluation metrics of the current model are calculated to determine if a new algorithm is needed.**3. Compare:** the new models are compared to each other to determine which model performs the best.**4. Rebuild:** the best performing model is re-built on the current state of data.

**How can a time-series data be declared as stationery?**

- The mean of the series should not be a function of time.

- The variance of the series should not be a function of time. This property is known as homoscedasticity.

- The covariance of the i th term and the (i+m) th term should not be a function of time.

**‘People who bought this also bought…’ recommendations seen on Amazon are a result of which algorithm?**

The recommendation engine is accomplished with **collaborative filtering**. Collaborative filtering explains the behavior of other users and their purchase history in terms of ratings, selection, etc.

The engine makes predictions on what might interest a person based on the preferences of other users. In this algorithm, item features are unknown.

For example, a sales page shows that a certain number of people buy a new phone and also buy tempered glass at the same time. Next time, when a person buys a phone, he or she may see a recommendation to buy tempered glass as well.

** What is a Generative Adversarial Network?**

Suppose there is a wine shop purchasing wine from dealers, which they resell later. But some dealers sell fake wine. In this case, the shop owner should be able to distinguish between fake and authentic wine. The forger will try different techniques to sell fake wine and make sure specific techniques go past the shop owner’s check. The shop owner would probably get some feedback from wine experts that some of the wine is not original. The owner would have to improve how he determines whether a wine is fake or authentic.

The forger’s goal is to create wines that are indistinguishable from the authentic ones while the shop owner intends to tell if the wine is real or not accurately.

• There is a noise vector coming into the forger who is generating fake wine.

• Here the forger acts as a Generator.

• The shop owner acts as a Discriminator.

• The Discriminator gets two inputs; one is the fake wine, while the other is the real authentic wine.

The shop owner has to figure out whether it is real or fake.

So, there are two primary components of Generative Adversarial Network (GAN) named:

1. Generator

2. Discriminator

The generator is a CNN that keeps keys producing images and is closer in appearance to the real images while the discriminator tries to determine the difference between real and fake images. The ultimate aim is to make the discriminator learn to identify real and fake images.

**You are given a dataset on cancer detection. You have built a classification model and achieved an accuracy of 96 percent. Why shouldn’t you be happy with your model performance? What can you do about it?**

Cancer detection results in imbalanced data. In an imbalanced dataset, accuracy should not be based as a measure of performance. It is important to focus on the remaining four percent, which represents the patients who were wrongly diagnosed. Early diagnosis is crucial when it comes to cancer detection and can greatly improve a patient’s prognosis.

Hence, to evaluate model performance, we should use Sensitivity (True Positive Rate), Specificity (True Negative Rate), F measure to determine the class wise performance of the classifier.

**We want to predict the probability of death from heart disease based on three risk factors: age, gender, and blood cholesterol level. What is the most appropriate algorithm for this case?**

The most appropriate algorithm for this case is l**ogistic regression**.

**After studying the behavior of a population, you have identified four specific individual types that are valuable to your study. You would like to find all users who are most similar to each individual type. Which algorithm is most appropriate for this study? **

As we are looking for grouping people together specifically by four different similarities, it indicates the value of k. Therefore, **K-means clustering** is the most appropriate algorithm for this study.

**You have run the association rules algorithm on your dataset, and the two rules {banana, apple} => {grape} and {apple, orange} => {grape} have been found to be relevant. What else must be true? **

**{grape, apple}** must be a frequent itemset.

**Your organization has a website where visitors randomly receive one of two coupons. It is also possible that visitors to the website will not receive a coupon. You have been asked to determine if offering a coupon to website visitors has any impact on their purchase decisions. Which analysis method should you use?**

**One-way ANOVA**: in statistics, one-way analysis of variance is a technique that can be used to compare means of two or more samples. This technique can be used only for numerical response data, the “Y”, usually one variable, and numerical or categorical input data, the “X”, always one variable, hence “oneway”.**The ANOVA tests the null hypothesis**, which states that samples in all groups are drawn from populations with the same mean values. To do this, two estimates are made of the population variance. **The ANOVA produces an F-statistic**, the ratio of the variance calculated among the means to the variance within the samples. If the group means are drawn from populations with the same mean values, the variance between the group means should be lower than the variance of the samples, following the central limit

theorem. A higher ratio therefore implies that the samples were drawn from populations with different mean values.

**What are the feature vectors?**

**A feature vector is an n-dimensional vector of numerical features that represent an object.** In machine learning, **feature vectors** are used to represent numeric or symbolic characteristics (called features) of an object in a mathematical way that’s easy to analyze.

**What is root cause analysis?**

Root cause analysis was initially developed to analyze industrial accidents but is now widely used in other areas. It is a **problem-solving technique used for isolating the root causes of faults or problems**. A factor is called a root cause if its deduction from the problem-fault-sequence averts the final undesirable event from recurring.

**Do gradient descent methods always converge to similar points?**

They do not, because in some cases, they reach a local minimum or a local optimum point. You would not reach the global optimum point. This is governed by the data and the starting conditions.

**What are the different Deep Learning Frameworks?**

• **PyTorch**: PyTorch is an open source machine learning library based on the Torch library, used for applications such as computer vision and natural language processing, primarily developed by Facebook’s AI Research lab. It is free and open-source software released under the Modified BSD license.

• **TensorFlow**: TensorFlow is a free and open-source software library for dataflow and differentiable programming across a range of tasks. It is a symbolic math library and is also used for machine learning applications such as neural networks. Licensed by Apache License 2.0. Developed by Google Brain Team.

• **Microsoft Cognitive Toolkit**: Microsoft Cognitive Toolkit describes neural networks as a series of computational steps via a directed graph.

• **Keras**: Keras is an open-source neural-network library written in Python. It is capable of running on top of TensorFlow, Microsoft Cognitive Toolkit, R, Theano, or PlaidML. Designed to enable fast experimentation with deep neural networks, it focuses on being user-friendly, modular, and extensible. Licensed by MIT.

**What are the different Deep Learning Frameworks?**

• **PyTorch**: PyTorch is an open source machine learning library based on the Torch library, used for applications such as computer vision and natural language processing, primarily developed by Facebook’s AI Research lab. It is free and open-source software released under the Modified BSD license.

• **TensorFlow**: TensorFlow is a free and open-source software library for dataflow and differentiable programming across a range of tasks. It is a symbolic math library and is also used for machine learning applications such as neural networks. Licensed by Apache License 2.0. Developed by Google Brain Team.

• **Microsoft Cognitive Toolkit**: Microsoft Cognitive Toolkit describes neural networks as a series of computational steps via a directed graph.

• **Keras**: Keras is an open-source neural-network library written in Python. It is capable of running on top of TensorFlow, Microsoft Cognitive Toolkit, R, Theano, or PlaidML. Designed to enable fast experimentation with deep neural networks, it focuses on being user-friendly, modular, and extensible. Licensed by MIT.

**How Does an LSTM Network Work?**

Reference: LTSM

Long-Short-Term Memory (LSTM) is a special kind of recurrent neural network capable of learning long-term dependencies, remembering information for long periods as its default behavior. There are three steps in an LSTM network:

• Step 1: The network decides what to forget and what to remember.

• Step 2: It selectively updates cell state values.

• Step 3: The network decides what part of the current state makes it to the output.

**What Is a Multi-layer Perceptron (MLP)?**

Reference: MLP

As in Neural Networks, MLPs have an input layer, a hidden layer, and an output layer. It has the same structure as a single layer perceptron with one or more hidden layers.

**Perceptron is a single layer neural network** and a multi-layer perceptron is called Neural Networks.

A (single layer) perceptron is a single layer neural network that works as a linear binary classifier. Being a single layer neural network, it can be trained without the use of more advanced algorithms like back propagation and instead can be trained by “stepping towards” your error in steps specified by a learning rate. When someone says perceptron, I usually think of the single layer version.

**What is exploding gradients? **

While training an RNN, if you see exponentially growing (very large) error gradients which accumulate and result in very large updates to neural network model weights during training, they’re known as exploding gradients. At an extreme, the values of weights can become so large as to overflow and result in NaN values. The explosion occurs through exponential growth by repeatedly multiplying gradients through the network layers that have values larger than 1.0.

This has the effect of your model is unstable and unable to learn from your training data.

There are some subtle signs that you may be suffering from exploding gradients during the training of your network, such as:

• The model is unable to get traction on your training data (e.g. poor loss).

• The model is unstable, resulting in large changes in loss from update to update.

• The model loss goes to NaN during training.

• The model weights quickly become very large during training.

• The error gradient values are consistently above 1.0 for each node and layer during training.

**Solutions****1. Re-Design the Network Model:**

a. In deep neural networks, exploding gradients may be addressed by redesigning the

network to have fewer layers. There may also be some benefit in using a smaller batch

size while training the network.

b. In RNNs, updating across fewer prior time steps during training, called truncated

Backpropagation through time, may reduce the exploding gradient problem.

**2. Use Long Short-Term Memory Networks:** In RNNs, exploding gradients can be reduced by using the Long Short-Term Memory (LSTM) memory units and perhaps related gated-type neuron structures. Adopting LSTM memory units is a new best practice for recurrent neural networks for sequence prediction.

**3. Use Gradient Clipping:** Exploding gradients can still occur in very deep Multilayer Perceptron networks with a large batch size and LSTMs with very long input sequence lengths. If exploding gradients are still occurring, you can check for and limit the size of gradients during the training of your network. This is called gradient clipping. Specifically, the values of the error gradient are checked against a threshold value and clipped or set to that threshold value if the error gradient exceeds the threshold.

**4. Use Weight Regularization:** another approach, if exploding gradients are still occurring, is to check the size of network weights and apply a penalty to the networks loss function for large weight values. This is called weight regularization and often an L1 (absolute weights) or an L2 (squared weights) penalty can be used.

**What is vanishing gradients? **

While training an RNN, your slope can become either too small; this makes the training difficult. When the slope is too small, the problem is known as a Vanishing Gradient. It leads to long training times, poor performance, and low accuracy.

• Hyperbolic tangent and Sigmoid/Soft-max suffer vanishing gradient.

• RNNs suffer vanishing gradient, LSTM no (so it is perfect to predict stock prices). In fact, the propagation of error through previous layers makes the gradient get smaller so the weights are not updated.

**Solutions**

1. **Choose RELU**

2. **Use LSTM (for RNNs)**

3. **Use ResNet (Residual Network)** → after some layers, add x again: F(x) → ⋯ → F(x) + x

4. **Multi-level hierarchy**: pre-train one layer at the time through unsupervised learning, then fine-tune via backpropagation

5. **Gradient checking**: debugging strategy used to numerically track and assess gradients during training.

**What is Gradient Descent?**

Let’s first explain what a gradient is. **A gradient is a mathematical function.** When calculated on a point of a function, it gives the hyperplane (or slope) of the directions in which the function increases more. The gradient vector can be interpreted as the “direction and rate of fastest increase”. If the gradient of a function is non-zero at a point p, the direction of the gradient is the direction in which the function increases most quickly from p, and the magnitude of the gradient is the rate of increase in that direction.

Further, the gradient is the zero vector at a point if and only if it is a stationary point (where the derivative vanishes).

In Data Science, it simply measures the change in all weights with regard to the change in error, as we are partially derivating by w the loss function.

**Gradient descent is a first-order iterative optimization algorithm for finding the minimum of a function.**

The goal of the gradient descent is to minimize a given function which, in our case, is the loss function of the neural network. To achieve this goal, it performs two steps iteratively.

1. Compute the slope (gradient) that is the first-order derivative of the function at the current point

2. Move-in the opposite direction of the slope increase from the current point by the computed amount

So, the idea is to pass the training set through the hidden layers of the neural network and then update the parameters of the layers by computing the gradients using the training samples from the training dataset.

Think of it like this. Suppose a man is at top of the valley and he wants to get to the bottom of the valley.

So, he goes down the slope. He decides his next position based on his current position and stops when he gets to the bottom of the valley which was his goal.

• Gradient descent is an iterative optimization algorithm that is popular and it is a base for many other optimization techniques, which tries to obtain minimal loss in a model by tuning the weights/parameters in the objective function.

• Types of Gradient Descent:

- Batch Gradient Descent
- Stochastic Gradient Descent
- Mini Batch Gradient Descent

• Steps to achieve minimal loss:

- The first stage in gradient descent is to pick a starting value (a starting point) for w1, which is set to 0 by many algorithms.
- The gradient descent algorithm then calculates the gradient of the loss curve at the starting point.
- The gradient always points in the direction of steepest increase in the loss function. The gradient descent algorithm takes a step in the direction of the negative gradient in order to reduce loss as quickly as possible.
- To determine the next point along the loss function curve, the gradient descent algorithm adds some fraction of the gradient’s magnitude to the starting point and moves forward.
- The gradient descent then repeats this process, edging ever closer to the minimum.

**What is vanishing gradients? **

While training an RNN, your slope can become either too small; this makes the training difficult. When the slope is too small, the problem is known as a Vanishing Gradient. It leads to long training times, poor performance, and low accuracy.

• Hyperbolic tangent and Sigmoid/Soft-max suffer vanishing gradient.

• RNNs suffer vanishing gradient, LSTM no (so it is perfect to predict stock prices). In fact, the propagation of error through previous layers makes the gradient get smaller so the weights are not updated.

**Solutions**

1. **Choose RELU**

2. **Use LSTM (for RNNs)**

3. **Use ResNet (Residual Network)** → after some layers, add x again: F(x) → ⋯ → F(x) + x

4. **Multi-level hierarchy**: pre-train one layer at the time through unsupervised learning, then fine-tune via backpropagation

5. **Gradient checking**: debugging strategy used to numerically track and assess gradients during training.

**What is Back Propagation and Explain it Works. **

Back propagation is a training algorithm used for neural network. In this method, we update the weights of each layer from the last layer recursively, with the formula:

It has the following steps:

• Forward Propagation of Training Data (initializing weights with random or pre-assigned values)

• Gradients are computed using output weights and target

• Back Propagate for computing gradients of error from output activation

• Update the Weights

**What are the variants of Back Propagation? **

Reference: Variants of back propagation

- Stochastic Gradient Descent: In Batch Gradient Descent we were considering all the examples for every step of Gradient Descent. But what if our dataset is very huge. Deep learning models crave for data. The more the data the more chances of a model to be good. Suppose our dataset has 5 million examples, then just to take one step the model will have to calculate the gradients of all the 5 million examples. This does not seem an efficient way. To tackle this problem, we have Stochastic Gradient Descent. In Stochastic Gradient Descent (SGD), we consider just one example at a time to take a single step. We do the following steps in one epoch for SGD:

1. Take an example

2. Feed it to Neural Network

3. Calculate its gradient

4. Use the gradient we calculated in step 3 to update the weights

5. Repeat steps 1–4 for all the examples in training dataset

Since we are considering just one example at a time the cost will fluctuate over the training examples and it will not necessarily decrease. But in the long run, you will see the cost decreasing with fluctuations. Also, because the cost is so fluctuating, it will never reach the minimum, but it will keep dancing around it. SGD can be used for larger datasets. It converges faster when the dataset is large as it causes updates to the parameters more frequently.

Stochastic Gradient Descent (SGD)

- Batch Gradient Descent: all the training data is taken into consideration to take a single step. We take the average of the gradients of all the training examples and then use that mean gradient to update our parameters. So that’s just one step of gradient descent in one epoch. Batch Gradient Descent is great for convex or relatively smooth error manifolds. In this case, we move somewhat directly towards an optimum solution. The graph of cost vs epochs is also quite smooth because we are averaging over all the gradients of training data for a single step. The cost keeps on decreasing over the epochs.

**Mini-batch Gradient Descent**: It’s one of the most popular optimization algorithms. It’s a variant of Stochastic Gradient Descent and here instead of single training example, mini batch of samples is used. Batch Gradient Descent can be used for smoother curves. SGD can be used when the dataset is large. Batch Gradient Descent converges directly to minima. SGD converges faster for larger datasets.

But, since in SGD we use only one example at a time, we cannot implement the vectorized implementation on it. This can slow down the computations. To tackle this problem, a mixture of Batch Gradient Descent and SGD is used. Neither we use all the dataset all at once nor we use the single example at a time. We use a batch of a fixed number of training examples which is less than the actual dataset and call it a mini-batch. Doing this helps us achieve the advantages of both the former variants we saw. So, after creating the mini-batches of fixed size, we do the following steps in one epoch:

1. Pick a mini-batch

2. Feed it to Neural Network

3. Calculate the mean gradient of the mini-batch

4. Use the mean gradient we calculated in step 3 to update the weights

5. Repeat steps 1–4 for the mini-batches we created

Just like SGD, the average cost over the epochs in mini-batch gradient descent fluctuates because we are averaging a small number of examples at a time. So, when we are using the mini-batch gradient descent we are updating our parameters frequently as well as we can use vectorized implementation for faster computations.

**While we continue to integrate ML systems in high-stakes environments such as medical settings, roads, command control centers, we need to ensure they do not cause the loss of life. How can you handle this?**

By focusing on the following, which includes everything outside of just developing SOTA models, as well inclusion of key stakeholders.

🔹Robustness: Create models that are resilient to adversaries, unusual situations, and Black Swan events

🔹Monitoring: Detect malicious use, monitor predictions, and discover unexpected model functionality

🔹Alignment: Build models that represent and safely optimize hard-to-specify human values

🔹External Safety: Use ML to address risks to how ML systems are handled, such as cyber attacks

Machine Learning Unsolved Problems_ n_SafetyDownload Unsolved Problems in ML Safety Here

**You are given a data set. The data set has missing values that spread along 1 standard deviation from the median. What percentage of data would remain unaffected? Why?**

Since the data is spread across the median, let’s assume it’s a normal distribution. We know, in a normal distribution, ~68% of the data lies in 1 standard deviation from mean (or mode, median), which leaves ~32% of the data unaffected. Therefore, ~32% of the data would remain unaffected by missing values.

**What are PCA, KPCA, and ICA used for?**

PCA (Principal Components Analysis), KPCA ( Kernel-based Principal Component Analysis) and ICA ( Independent Component Analysis) are important feature extraction techniques used for dimensionality reduction.

**What is the bias-variance decomposition of classification error in the ensemble method?**

The expected error of a learning algorithm can be decomposed into bias and variance. A bias term measures how closely the average classifier produced by the learning algorithm matches the target function. The variance term measures how much the learning algorithm’s prediction fluctuates for different training sets.

**When is Ridge regression favorable over Lasso regression?**

You can quote ISLR’s authors Hastie, Tibshirani who asserted that, in the presence of few variables with medium / large sized effect, use lasso regression. In presence of many variables with small/medium-sized effects, use ridge regression.

Conceptually, we can say, lasso regression (L1) does both variable selection and parameter shrinkage, whereas Ridge regression only does parameter shrinkage and end up including all the coefficients in the model. In the presence of correlated variables, ridge regression might be the preferred choice. Also, ridge regression works best in situations where the least square estimates have higher variance. Therefore, it depends on our model objective.

**You’ve built a random forest model with 10000 trees. You got delighted after getting training error as 0.00. But, the validation error is 34.23. What is going on? Haven’t you trained your model perfectly?**

The model has overfitted. Training error 0.00 means the classifier has mimicked the training data patterns to an extent, that they are not available in the unseen data. Hence, when this classifier was run on an unseen sample, it couldn’t find those patterns and returned predictions with higher error. In a random forest, it happens when we use a larger number of trees than necessary. Hence, to avoid this situation, we should tune the number of trees using cross-validation.

**What is a convex hull?**

In the case of linearly separable data, the convex hull represents the outer boundaries of the two groups of data points. Once the convex hull is created, we get maximum margin hyperplane (MMH) as a perpendicular bisector between two convex hulls. MMH is the line which attempts to create the greatest separation between two groups.

**What do you understand by Type I vs Type II error?**

Type I error is committed when the null hypothesis is true and we reject it, also known as a ‘False Positive’. Type II error is committed when the null hypothesis is false and we accept it, also known as ‘False Negative’.

In the context of the confusion matrix, we can say Type I error occurs when we classify a value as positive (1) when it is actually negative (0). Type II error occurs when we classify a value as negative (0) when it is actually positive(1).

**In k-means or kNN, we use euclidean distance to calculate the distance between nearest neighbors. Why not manhattan distance?**

We don’t use manhattan distance because it calculates distance horizontally or vertically only. It has dimension restrictions. On the other hand, the euclidean metric can be used in any space to calculate distance. Since the data points can be present in any dimension, euclidean distance is a more viable option.

**Example:** Think of a chessboard, the movement made by a bishop or a rook is calculated by manhattan distance because of their respective vertical & horizontal movements.

**Do you suggest that treating a categorical variable as a continuous variable would result in a better predictive model?**

For better predictions, the categorical variable can be considered as a continuous variable only when the variable is ordinal in nature.

**OLS is to linear regression what the maximum likelihood is logistic regression. Explain the statement.**

OLS and Maximum likelihood are the methods used by the respective regression methods to approximate the unknown parameter (coefficient) value. In simple words, Ordinary least square(OLS) is a method used in linear regression which approximates the parameters resulting in minimum distance between actual and predicted values. Maximum

Likelihood helps in choosing the values of parameters which maximizes the likelihood that the parameters are most likely to produce observed data.

**When does regularization becomes necessary in Machine Learning?**

Regularization becomes necessary when the model begins to overfit/underfit. This technique introduces a cost term for bringing in more features with the objective function. Hence, it tries to push the coefficients for many variables to zero and hence reduce the cost term. This helps to reduce model complexity so that the model can become better at predicting (generalizing).

**What is Linear Regression?**

Linear Regression is a supervised Machine Learning algorithm. It is used to find the linear relationship between the dependent and the independent variables for predictive analysis.

• Linear regression assumes that the relationship between the features and the target vector is approximately linear. That is, the effect of the features on the target vector is constant.

• In linear regression, the target variable y is assumed to follow a linear function of one or more predictor variables plus some random error. The machine learning task is to estimate the parameters of this equation which can be achieved in two ways:

• The first approach is through the lens of minimizing loss. A common practice in machine learning is to choose a loss function that defines how well a model with a given set of parameters estimates the observed data. The most common loss function for linear regression is squared error loss.

• The second approach is through the lens of maximizing the likelihood. Another common practice in machine learning is to model the target as a random variable whose distribution depends on one or more parameters, and then find the parameters that maximize its likelihood.

Credit: Vikram K.

**What is the Variance Inflation Factor?**

Variance Inflation Factor (VIF) is the estimate of the volume of multicollinearity in a collection of many regression variables.

VIF = Variance of the model / Variance of the model with a single independent variable

We have to calculate this ratio for every independent variable. If VIF is high, then it shows the high collinearity of the independent variables.

**We know that one hot encoding increases the dimensionality of a dataset, but label encoding doesn’t. How?**

When we use one-hot encoding, there is an increase in the dimensionality of a dataset. The reason for the increase in dimensionality is that, for every class in the categorical variables, it forms a different variable.

**What is a Decision Tree?**

A decision tree is used to explain the sequence of actions that must be performed to get the desired output. It is a hierarchical diagram that shows the actions.

**What is the Binarizing of data? How to Binarize?**

In most of the Machine Learning Interviews, apart from theoretical questions, interviewers focus on the implementation part. So, this ML Interview Questions focused on the implementation of the theoretical concepts.

Converting data into binary values on the basis of threshold values is known as the binarizing of data. The values that are less than the threshold are set to 0 and the values that are greater than the threshold are set to 1.

This process is useful when we have to perform feature engineering, and we can also use it for adding unique features.

**What is cross-validation?**

Cross-validation is essentially a technique used to assess how well a model performs on a new independent dataset. The simplest example of cross-validation is when you split your data into two groups: training data and testing data, where you use the training data to build the model and the testing data to test the model.

• Cross-validation is a resampling procedure used to evaluate machine learning models on a limited data sample. The procedure has a single parameter called k that refers to the number of groups that a given data sample is to be split into. As such, the procedure is often called k-fold cross-validation.

• Cross-validation is primarily used in applied machine learning to estimate the skill of a machine learning model on unseen data. That is, to use a limited sample in order to estimate how the model is expected to perform in general when used to make predictions on data not used during the training of the model.

• It is a popular method because it is simple to understand and because it generally results in a less biased or less optimistic estimate of the model skill than other methods, such as a simple train/test split.

• Procedure for K-Fold Cross Validation:

1. Shuffle the dataset randomly.

2. Split the dataset into k groups

3. For each unique group:

a. Take the group as a holdout or test data set

b. Take the remaining groups as a training data set

c. Fit a model on the training set and evaluate it on the test set

d. Retain the evaluation score and discard the model

4. Summarize the skill of the model using the sample of model evaluation scores

Credit: Vikram K.

**When would you use random forests Vs SVM and why?**

There are a couple of reasons why a random forest is a better choice of the model than a support vector machine:

● Random forests allow you to determine the feature importance. SVM’s can’t do this.

● Random forests are much quicker and simpler to build than an SVM.

● For multi-class classification problems, SVMs require a one-vs-rest method, which is less scalable and more memory intensive.

**What are the drawbacks of a linear model?**

There are a couple of drawbacks of a linear model:

● A linear model holds some strong assumptions that may not be true in the application. It assumes a linear relationship, multivariate normality, no or little multicollinearity, no auto-correlation, and homoscedasticity

● A linear model can’t be used for discrete or binary outcomes.

● You can’t vary the model flexibility of a linear model.

**While we continue to integrate ML systems in high-stakes environments such as medical settings, roads, command control centers, we need to ensure they do not cause the loss of life. How can you handle this?**

By focusing on the following, which includes everything outside of just developing SOTA models, as well inclusion of key stakeholders.

🔹Robustness: Create models that are resilient to adversaries, unusual situations, and Black Swan events

🔹Monitoring: Detect malicious use, monitor predictions, and discover unexpected model functionality

🔹Alignment: Build models that represent and safely optimize hard-to-specify human values

🔹External Safety: Use ML to address risks to how ML systems are handled, such as cyber attacks

Machine Learning Unsolved Problems_ n_SafetyDownload Unsolved Problems in ML Safety Here

**You are given a data set. The data set has missing values that spread along 1 standard deviation from the median. What percentage of data would remain unaffected? Why?**

Since the data is spread across the median, let’s assume it’s a normal distribution. We know, in a normal distribution, ~68% of the data lies in 1 standard deviation from mean (or mode, median), which leaves ~32% of the data unaffected. Therefore, ~32% of the data would remain unaffected by missing values.

**What are PCA, KPCA, and ICA used for?**

PCA (Principal Components Analysis), KPCA ( Kernel-based Principal Component Analysis) and ICA ( Independent Component Analysis) are important feature extraction techniques used for dimensionality reduction.

**What are support vector machines?**

Support vector machines are supervised learning algorithms used for classification and regression analysis.

**What is batch statistical learning?**

Statistical learning techniques allow learning a function or predictor from a set of observed data that can make predictions about unseen or future data. These techniques provide guarantees on the performance of the learned predictor on the future unseen data based on a statistical assumption on the data generating process.

**What is the bias-variance decomposition of classification error in the ensemble method?**

The expected error of a learning algorithm can be decomposed into bias and variance. A bias term measures how closely the average classifier produced by the learning algorithm matches the target function. The variance term measures how much the learning algorithm’s prediction fluctuates for different training sets.

**When is Ridge regression favorable over Lasso regression?**

You can quote ISLR’s authors Hastie, Tibshirani who asserted that, in the presence of few variables with medium / large sized effect, use lasso regression. In presence of many variables with small/medium-sized effects, use ridge regression.

Conceptually, we can say, lasso regression (L1) does both variable selection and parameter shrinkage, whereas Ridge regression only does parameter shrinkage and end up including all the coefficients in the model. In the presence of correlated variables, ridge regression might be the preferred choice. Also, ridge regression works best in situations where the least square estimates have higher variance. Therefore, it depends on our model objective.

**You’ve built a random forest model with 10000 trees. You got delighted after getting training error as 0.00. But, the validation error is 34.23. What is going on? Haven’t you trained your model perfectly?**

The model has overfitted. Training error 0.00 means the classifier has mimicked the training data patterns to an extent, that they are not available in the unseen data. Hence, when this classifier was run on an unseen sample, it couldn’t find those patterns and returned predictions with higher error. In a random forest, it happens when we use a larger number of trees than necessary. Hence, to avoid this situation, we should tune the number of trees using cross-validation.

**What is a convex hull?**

In the case of linearly separable data, the convex hull represents the outer boundaries of the two groups of data points. Once the convex hull is created, we get maximum margin hyperplane (MMH) as a perpendicular bisector between two convex hulls. MMH is the line which attempts to create the greatest separation between two groups.

**What do you understand by Type I vs Type II error?**

Type I error is committed when the null hypothesis is true and we reject it, also known as a ‘False Positive’. Type II error is committed when the null hypothesis is false and we accept it, also known as ‘False Negative’.

In the context of the confusion matrix, we can say Type I error occurs when we classify a value as positive (1) when it is actually negative (0). Type II error occurs when we classify a value as negative (0) when it is actually positive(1).

**In k-means or kNN, we use euclidean distance to calculate the distance between nearest neighbors. Why not manhattan distance?**

We don’t use manhattan distance because it calculates distance horizontally or vertically only. It has dimension restrictions. On the other hand, the euclidean metric can be used in any space to calculate distance. Since the data points can be present in any dimension, euclidean distance is a more viable option.

**Example:** Think of a chessboard, the movement made by a bishop or a rook is calculated by manhattan distance because of their respective vertical & horizontal movements.

**Do you suggest that treating a categorical variable as a continuous variable would result in a better predictive model?**

For better predictions, the categorical variable can be considered as a continuous variable only when the variable is ordinal in nature.

**OLS is to linear regression wha the maximum likelihood is logistic regression. Explain the statement.**

OLS and Maximum likelihood are the methods used by the respective regression methods to approximate the unknown parameter (coefficient) value. In simple words, Ordinary least square(OLS) is a method used in linear regression which approximates the parameters resulting in minimum distance between actual and predicted values. Maximum

Likelihood helps in choosing the values of parameters which maximizes the likelihood that the parameters are most likely to produce observed data.

**When does regularization becomes necessary in Machine Learning?**

Regularization becomes necessary when the model begins to overfit/underfit. This technique introduces a cost term for bringing in more features with the objective function. Hence, it tries to push the coefficients for many variables to zero and hence reduce the cost term. This helps to reduce model complexity so that the model can become better at predicting (generalizing).

**What is Linear Regression?**

Linear Regression is a supervised Machine Learning algorithm. It is used to find the linear relationship between the dependent and the independent variables for predictive analysis.

**What is the Variance Inflation Factor?**

Variance Inflation Factor (VIF) is the estimate of the volume of multicollinearity in a collection of many regression variables.

VIF = Variance of the model / Variance of the model with a single independent variable

We have to calculate this ratio for every independent variable. If VIF is high, then it shows the high collinearity of the independent variables.

**We know that one hot encoding increases the dimensionality of a dataset, but label encoding doesn’t. How?**

When we use one-hot encoding, there is an increase in the dimensionality of a dataset. The reason for the increase in dimensionality is that, for every class in the categorical variables, it forms a different variable.

**What is a Decision Tree?**

A decision tree is used to explain the sequence of actions that must be performed to get the desired output. It is a hierarchical diagram that shows the actions.

**What is the Binarizing of data? How to Binarize?**

In most of the Machine Learning Interviews, apart from theoretical questions, interviewers focus on the implementation part. So, this ML Interview Questions focused on the implementation of the theoretical concepts.

Converting data into binary values on the basis of threshold values is known as the binarizing of data. The values that are less than the threshold are set to 0 and the values that are greater than the threshold are set to 1.

This process is useful when we have to perform feature engineering, and we can also use it for adding unique features.

**What is cross-validation?**

Cross-validation is essentially a technique used to assess how well a model performs on a new independent dataset. The simplest example of cross-validation is when you split your data into two groups: training data and testing data, where you use the training data to build the model and the testing data to test the model.

**When would you use random forests Vs SVM and why?**

There are a couple of reasons why a random forest is a better choice of the model than a support vector machine:

● Random forests allow you to determine the feature importance. SVM’s can’t do this.

● Random forests are much quicker and simpler to build than an SVM.

● For multi-class classification problems, SVMs require a one-vs-rest method, which is less scalable and more memory intensive.

**What are the drawbacks of a linear model?**

There are a couple of drawbacks of a linear model:

● A linear model holds some strong assumptions that may not be true in the application. It assumes a linear relationship, multivariate normality, no or little multicollinearity, no auto-correlation, and homoscedasticity

● A linear model can’t be used for discrete or binary outcomes.

● You can’t vary the model flexibility of a linear model.

**Do you think 50 small decision trees are better than a large one? Why?**

Another way of asking this question is “Is a random forest a better model than a decision tree?”

And the answer is yes because a random forest is an ensemble method that takes many weak decision trees to make a strong learner. Random forests are more accurate, more robust, and less prone to overfitting.

**What is a kernel? Explain the kernel trick**

A kernel is a way of computing the dot product of two vectors x and y in some (possibly very high dimensional) feature space, which is why kernel functions are sometimes called “generalized dot product”

The kernel trick is a method of using a linear classifier to solve a non-linear problem by transforming linearly inseparable data to linearly separable ones in a higher dimension.

**State the differences between causality and correlation?**

Causality applies to situations where one action, say X, causes an outcome, say Y, whereas Correlation is just relating one action (X) to another action(Y) but X does not necessarily cause Y.

**What is the exploding gradient problem while using the backpropagation technique?**

When large error gradients accumulate and result in large changes in the neural network weights during training, it is called the exploding gradient problem. The values of weights can become so large as to overflow and result in NaN values. This makes the model unstable and the learning of the model to stall just like the vanishing gradient problem.

**What do you mean by Associative Rule Mining (ARM)?**

Associative Rule Mining is one of the techniques to discover patterns in data like features (dimensions) which occur together and features (dimensions) which are correlated.

**What is Marginalization? Explain the process.**

Marginalization is summing the probability of a random variable X given the joint probability distribution of X with other variables. It is an application of the law of total probability.

**Why is the rotation of components so important in Principle Component Analysis(PCA)?**

Rotation in PCA is very important as it maximizes the separation within the variance obtained by all the components because of which interpretation of components would become easier. If the components are not rotated, then we need extended components to describe the variance of the components.

**What is the difference between regularization and normalization?**

Normalization adjusts the data; regularization adjusts the prediction function. If your data is on very different scales (especially low to high), you would want to normalize the data. Alter each column to have compatible basic statistics. This can be helpful to make sure there is no loss of accuracy. One of the goals of model training is to identify the signal and ignore the noise if the model is given free rein to minimize error, there is a possibility of suffering from overfitting.

Regularization imposes some control on this by providing simpler fitting functions over complex ones.

**How does the SVM algorithm deal with self-learning?**

SVM has a learning rate and expansion rate which takes care of this. The learning rate compensates or penalizes the hyperplanes for making all the wrong moves and expansion rate deals with finding the maximum separation area between classes.

**How do you handle outliers in the data?**

Outlier is an observation in the data set that is far away from other observations in the data set.

We can discover outliers using tools and functions like box plot, scatter plot, Z-Score, IQR score etc. and then handle them based on the visualization we have got. To handle outliers, we can cap at some threshold, use transformations to reduce skewness of the data and remove outliers if they are anomalies or errors.

**What are some techniques used to find similarities in the recommendation system?**

Pearson correlation and Cosine correlation are techniques used to find similarities in recommendation systems.

**Why would you Prune your tree?**

In the context of data science or AIML, pruning refers to the process of reducing redundant branches of a decision tree. Decision Trees are prone to overfitting, pruning the tree helps to reduce the size and minimizes the chances of overfitting. Pruning involves turning branches of a decision tree into leaf nodes and removing the leaf nodes from the original branch. It serves as a tool to perform the tradeoff.

**What are some of the EDA Techniques?**

Exploratory Data Analysis (EDA) helps analysts to understand the data better and forms the foundation of better models.

Visualization

● Univariate visualization

● Bivariate visualization

● Multivariate visualization

Missing Value Treatment – Replace missing values with Either Mean/Median Outlier Detection – Use Boxplot to identify the distribution of Outliers, then Apply IQR to set the boundary for IQR

**What is data augmentation?**

Data augmentation is a technique for synthesizing new data by modifying existing data in such a way that the target is not changed, or it is changed in a known way.

CV is one of the fields where data augmentation is very useful. There are many modifications that we can do to images:

● Resize

● Horizontal or vertical flip

● Rotate

● Add noise

● Deform

● Modify colors

Each problem needs a customized data augmentation pipeline. For example, on OCR, doing flips will change the text and won’t be beneficial; however, resizes and small rotations may **help.**

**What is Inductive Logic Programming in Machine Learning (ILP)?**

Inductive Logic Programming (ILP) is a subfield of machine learning which uses logic programming representing background knowledge and examples.

**What is the difference between inductive machine learning and deductive machine learning?**

The difference between inductive machine learning and deductive machine learning are as follows: machine-learning where the model learns by examples from a set of observed instances to draw a generalized conclusion whereas in deductive learning the model first draws the conclusion and then the conclusion is drawn.

**What is the Difference between machine learning and deep learning?**

Machine learning is a branch of computer science and a method to implement artificial intelligence. This technique provides the ability to automatically learn and improve from experiences without being explicitly programmed.

Deep learning can be said as a subset of machine learning. It is mainly based on the artificial neural network where data is taken as an input and the technique makes intuitive decisions using the artificial neural network.

**What Are The Steps Involved In Machine Learning Project?**

As you plan for doing a machine learning project. There are several important steps you must follow to achieve a good working model and they are **data collection, data preparation, choosing a machine learning model, training the model, model evaluation, parameter tuning and lastly prediction.**

**What are Differences between Artificial Intelligence and Machine Learning?**

Artificial intelligence is a broader prospect than machine learning. Artificial intelligence mimics the cognitive functions of the human brain. The purpose of AI is to carry out a task in an intelligent manner based on algorithms. On the other hand, machine learning is a subclass of artificial intelligence. To develop an autonomous machine in such a way so that it can learn without being explicitly programmed is the goal of machine learning.

**What are the steps Needed to choose the Appropriate Machine Learning Algorithm for your Classification problem?**

Firstly, you need to have a clear picture of your data, your constraints, and your problems before heading towards different machine learning algorithms. Secondly, you have to understand which type and kind of data you have because it plays a primary role in deciding which algorithm you have to use.

Following this step is the data categorization step, which is a two-step process – categorization by input and categorization by output. The next step is to understand your constraints; that is, what is your data storage capacity? How fast the prediction has to be? etc.

Finally, find the available machine learning algorithms and implement them wisely. Along with that, also try to optimize the hyperparameters which can be done in three ways – grid search, random search, and Bayesian optimization.

**What is the Convex Function?**

A convex function is a continuous function, and the value of the midpoint at every interval in its given domain is less than the numerical mean of the values at the two ends of the interval.

**What’s the Relationship between True Positive Rate and Recall?**

The True positive rate in machine learning is the percentage of the positives that have been properly acknowledged, and recall is just the count of the results that have been correctly identified and are relevant. Therefore, they are the same things, just having different names. It is also known as sensitivity.

**What are some tools for parallelizing Machine Learning Algorithms?**

Almost all machine learning algorithms are easy to serialize. Some of the basic tools for parallelizing are Matlab, Weka, R, Octave, or the Python-based sci-kit learn.

**What is meant by Genetic Programming?**

Genetic Programming (GP) is almost similar to an Evolutionary Algorithm, a subset of machine learning. Genetic programming software systems implement an algorithm that uses random mutation, a fitness function, crossover, and multiple generations of evolution to resolve a user-defined task. The genetic programming model is based on testing and choosing the best option among a set of results.

**What is meant by Bayesian Networks?**

Bayesian Networks also referred to as ‘belief networks’ or ‘casual networks’, are used to represent the graphical model for probability relationship among a set of variables.

For example, a Bayesian network can be used to represent the probabilistic relationships between diseases and symptoms. As per the symptoms, the network can also compute the probabilities of the presence of various diseases.

Efficient algorithms can perform inference or learning in Bayesian networks. Bayesian networks which relate the variables (e.g., speech signals or protein sequences) are called dynamic Bayesian networks.

**Which are the two components of the Bayesian logic program?**

A Bayesian logic program consists of two components:

● Logical It contains a set of Bayesian Clauses, which capture the qualitative structure of the domain.

● Quantitative It is used to encode quantitative information about the domain.

**How is machine learning used in day-to-day life?**

Most of the people are already using machine learning in their everyday life. Assume that you are engaging with the internet, you are actually expressing your preferences, likes, dislikes through your searches. All these things are picked up by cookies coming on your computer, from this, the behavior of a user is evaluated. It helps to increase the progress of a user through the internet and provide similar suggestions.

The navigation system can also be considered as one of the examples where we are using machine learning to calculate a distance between two places using optimization techniques.

**What is Sampling. Why do we need it?**

Sampling is a process of choosing a subset from a target population that would serve as its representative. We use the data from the sample to understand the pattern in the community as a whole. Sampling is necessary because often, we can not gather or process the complete data within a reasonable time.

**What does the term decision boundary mean?**

A decision boundary or a decision surface is a hypersurface which divides the underlying feature space into two subspaces, one for each class. If the decision boundary is a hyperplane, then the classes are linearly separable.

**Define entropy?**

Entropy is the measure of uncertainty associated with random variable Y. It is the expected number of bits required to communicate the value of the variable.

**Indicate the top intents of machine learning?**

The top intents of machine learning are stated below,

● The system gets information from the already established computations to give well-founded decisions and outputs.

● It locates certain patterns in the data and then makes certain predictions on it to provide answers on matters.

**Highlight the differences between the Generative model and the Discriminative model?**

The aim of the Generative model is to generate new samples from the same distribution and new data instances, Whereas, the Discriminative model highlights the differences between different kinds of data instances. It tries to learn directly from the data and then classifies the data.

**Identify the most important aptitudes of a machine learning engineer?**

Machine learning allows the computer to learn itself without being decidedly programmed. It helps the system to learn from experience and then improve from its mistakes. The intelligence system, which is based on machine learning, can learn from recorded data and past incidents.

In-depth knowledge of statistics, probability, data modelling, programming language, as well as CS, Application of ML Libraries and algorithms, and software design is required to become a successful machine learning engineer.

**What is feature engineering? How do you apply it in the process of modelling?**

Feature engineering is the process of transforming raw data into features that better represent the underlying problem to the predictive models, resulting in improved model accuracy on unseen data.

**How can learning curves help create a better model?**

Learning curves give the indication of the presence of overfitting or underfitting. In a learning curve, the training error and cross-validating error are plotted against the number of training data points.

**Perception: Vision, Audio, Speech, Natural Language**

**NLP: TF-IDF helps you to establish what?**

**List 10 use cases to be solved using NLP techniques?**

● Sentiment Analysis

● Language Translation (English to German, Chinese to English, etc..)

● Document Summarization

● Question Answering

● Sentence Completion

● Attribute extraction (Key information extraction from the documents)

● Chatbot interactions

● Topic classification

● Intent extraction

● Grammar or Sentence correction

● Image captioning

● Document Ranking

● Natural Language inference

**Which NLP model gives the best accuracy amongst the following: BERT, XLNET, GPT-2, ELMo**

XLNET has given best accuracy amongst all the models. It has outperformed BERT on 20 tasks and achieves state of art results on 18 tasks including sentiment analysis, question answering, natural language inference, etc.

**What is Naive Bayes algorithm, When we can use this algorithm in NLP?**

Naive Bayes algorithm is a collection of classifiers which works on the principles of the Bayes’theorem. This series of NLP model forms a family of algorithms that can be used for a wide range of classification tasks including sentiment prediction, filtering of spam, classifying documents and more.

Naive Bayes algorithm converges faster and requires less training data. Compared to other discriminative models like logistic regression, Naive Bayes model takes lesser time to train. This algorithm is perfect for use while working with multiple classes and text classification where the data is dynamic and changes frequently.

**Explain Dependency Parsing in NLP?**

Dependency Parsing, also known as Syntactic parsing in NLP is a process of assigning syntactic structure to a sentence and identifying its dependency parses. This process is crucial to understand the correlations between the “head” words in the syntactic structure.

The process of dependency parsing can be a little complex considering how any sentence can have more than one dependency parses. Multiple parse trees are known as ambiguities.

Dependency parsing needs to resolve these ambiguities in order to effectively assign a syntactic structure to a sentence.

Dependency parsing can be used in the semantic analysis of a sentence apart from the syntactic structuring.

**What is text Summarization?**

Text summarization is the process of shortening a long piece of text with its meaning and effect intact. Text summarization intends to create a summary of any given piece of text and outlines the main points of the document. This technique has improved in recent times and is capable of summarizing volumes of text successfully.

Text summarization has proved to a blessing since machines can summarize large volumes of text in no time which would otherwise be really time-consuming. There are two types of text summarization:

● Extraction-based summarization

● Abstraction-based summarization

**What is NLTK? How is it different from Spacy?**

NLTK or Natural Language Toolkit is a series of libraries and programs that are used for symbolic and statistical natural language processing. This toolkit contains some of the most powerful libraries that can work on different ML techniques to break down and understand human language. NLTK is used for Lemmatization, Punctuation, Character count, Tokenization, and Stemming.

The difference between NLTK and Spacey are as follows:

● While NLTK has a collection of programs to choose from, Spacey contains only the best suited algorithm for a problem in its toolkit

● NLTK supports a wider range of languages compared to Spacey (Spacey supports only 7 languages)

● While Spacey has an object-oriented library, NLTK has a string processing library

● Spacey can support word vectors while NLTK cannot

**What is information extraction?**

Information extraction in the context of Natural Language Processing refers to the technique of extracting structured information automatically from unstructured sources to ascribe meaning to it. This can include extracting information regarding attributes of entities, relationship between different entities and more. The various models of information extraction includes:

● Tagger Module

● Relation Extraction Module

● Fact Extraction Module

● Entity Extraction Module

● Sentiment Analysis Module

● Network Graph Module

● Document Classification & Language Modeling Module

**What is Bag of Words?**

Bag of Words is a commonly used model that depends on word frequencies or occurrences to train a classifier. This model creates an occurrence matrix for documents or sentences irrespective of its grammatical structure or word order.

**What is Pragmatic Ambiguity in NLP?**

Pragmatic ambiguity refers to those words which have more than one meaning and their use in any sentence can depend entirely on the context. Pragmatic ambiguity can result in multiple interpretations of the same sentence. More often than not, we come across sentences which have words with multiple meanings, making the sentence open to interpretation. This multiple interpretation causes ambiguity and is known as Pragmatic ambiguity in NLP.

**What is a Masked Language Model?**

Masked language models help learners to understand deep representations in downstream tasks by taking an output from the corrupt input. This model is often used to predict the words to be used in a sentence.

**What are the best NLP Tools?**

Some of the best NLP tools from open sources are:

● SpaCy

● TextBlob

● Textacy

● Natural language Toolkit

● Retext

● NLP.js

● Stanford NLP

● CogcompNLP

**What is POS tagging?**

Parts of speech tagging better known as POS tagging refers to the process of identifying specific words in a document and group them as part of speech, based on its context. POS tagging is also known as grammatical tagging since it involves understanding grammatical structures and identifying the respective component.

POS tagging is a complicated process since the same word can be different parts of speech depending on the context. The same generic process used for word mapping is quite ineffective for POS tagging because of the same reason.

**What is NES?**

Name entity recognition is more commonly known as NER is the process of identifying specific entities in a text document which are more informative and have a unique context. These often denote places, people, organizations, and more. Even though it seems like these entities are proper nouns, the NER process is far from identifying just the nouns. In fact, NER involves entity

chunking or extraction wherein entities are segmented to categorize them under different predefined classes. This step further helps in extracting information.

**Explain the Masked Language Model?**

Masked language modelling is the process in which the output is taken from the corrupted input.

This model helps the learners to master the deep representations in downstream tasks. You can predict a word from the other words of the sentence using this model.

**What is pragmatic analysis in NLP?**

Pragmatic Analysis: It deals with outside word knowledge, which means knowledge that is external to the documents and/or queries. Pragmatics analysis that focuses on what was described is reinterpreted by what it actually meant, deriving the various aspects of language that require real-world knowledge.

**What is perplexity in NLP?**

The word “perplexed” means “puzzled” or “confused”, thus Perplexity in general means the inability to tackle something complicated and a problem that is not specified. Therefore, Perplexity in NLP is a way to determine the extent of uncertainty in predicting some text.

In NLP, perplexity is a way of evaluating language models. Perplexity can be high and low; Low perplexity is ethical because the inability to deal with any complicated problem is less while high perplexity is terrible because the failure to deal with a complicated is high.

**What is ngram in NLP?**

N-gram in NLP is simply a sequence of n words, and we also conclude the sentences which appeared more frequently, for example, let us consider the progression of these three words:

● New York (2 gram)

● The Golden Compass (3 gram)

● She was there in the hotel (4 gram)

Now from the above sequence, we can easily conclude that sentence (a) appeared more frequently than the other two sentences, and the last sentence(c) is not seen that often. Now if we assign probability in the occurrence of an n-gram, then it will be advantageous. It would help in making next-word predictions and in spelling error corrections.

Explain differences between AI, Machine Learning and NLP

**Why self-attention is awesome?**

“In terms of computational complexity, self-attention layers are faster than recurrent layers when the sequence length n is smaller than the representation dimensionality d, which is most often the case with sentence representations used by state-of-the-art models in machine translations, such as word-piece and byte-pair representations.” — from Attention is all you need.

**What are stop words?**

Stop words are said to be useless data for a search engine. Words such as articles, prepositions, etc. are considered as stop words. There are stop words such as was, were, is, am, the, a, an, how, why, and many more. In Natural Language Processing, we eliminate the stop words to understand and analyze the meaning of a sentence. The removal of stop words is one of the most important tasks for search engines. Engineers design the algorithms of search engines in such a way that they ignore the use of stop words. This helps show the relevant search result for a query.

**What is Latent Semantic Indexing (LSI)?**

Latent semantic indexing is a mathematical technique used to improve the accuracy of the information retrieval process. The design of LSI algorithms allows machines to detect the hidden (latent) correlation between semantics (words). To enhance information understanding, machines generate various concepts that associate with the words of a sentence.

The technique used for information understanding is called singular value decomposition. It is generally used to handle static and unstructured data. The matrix obtained for singular value decomposition contains rows for words and columns for documents. This method best suits to identify components and group them according to their types.

The main principle behind LSI is that words carry a similar meaning when used in a similar context.

Computational LSI models are slow in comparison to other models. However, they are good at contextual awareness that helps improve the analysis and understanding of a text or a document.

**What are Regular Expressions?**

A regular expression is used to match and tag words. It consists of a series of characters for matching strings.

Suppose, if A and B are regular expressions, then the following are true for them:

● If {ɛ} is a regular language, then ɛ is a regular expression for it.

● If A and B are regular expressions, then A + B is also a regular expression within the language {A, B}.

● If A and B are regular expressions, then the concatenation of A and B (A.B) is a regular expression.

● If A is a regular expression, then A* (A occurring multiple times) is also a regular expression.

**What are unigrams, bigrams, trigrams, and n-grams in NLP?**

When we parse a sentence one word at a time, then it is called a unigram. The sentence parsed two words at a time is a bigram.

When the sentence is parsed three words at a time, then it is a trigram. Similarly, n-gram refers to the parsing of n words at a time.

**What are the steps involved in solving an NLP problem?**

Below are the steps involved in solving an NLP problem:

1. Gather the text from the available dataset or by web scraping

2. Apply stemming and lemmatization for text cleaning

3. Apply feature engineering techniques

4. Embed using word2vec

5. Train the built model using neural networks or other Machine Learning techniques

6. Evaluate the model’s performance

7. Make appropriate changes in the model

8. Deploy the model

**There have some various common elements of natural language processing. Those elements are very important for understanding NLP properly, can you please explain the same in details with an example?**

There have a lot of components normally using by natural language processing (NLP). Some of the major components are explained below:

● Extraction of Entity: It actually identifying and extracting some critical data from the available information which help to segmentation of provided sentence on identifying each entity. It can help in identifying one human that it’s fictional or real, same kind of reality identification for any organization, events or any geographic location etc.

● The analysis in a syntactic way: it mainly helps for maintaining ordering properly of the available words.

**In the case of processing natural language, we normally mentioned one common terminology NLP and binding every language with the same terminology properly. Please explain in details about this NLP terminology with an example?**

This is the basic NLP Interview Questions asked in an interview. There have some several factors available in case of explaining natural language processing. Some of the key factors are given below:

● Vectors and Weights: Google Word vectors, length of TF-IDF, varieties documents, word vectors, TF-IDF.

● Structure of Text: Named Entities, tagging of part of speech, identifying the head of the sentence.

● Analysis of sentiment: Know about the features of sentiment, entities available for the sentiment, sentiment common dictionary.

● Classification of Text: Learning supervising, set off a train, set of validation in Dev, Set of define test, a feature of the individual text, LDA.

● Reading of Machine Language: Extraction of the possible entity, linking with an individual entity, DBpedia, some libraries like Pikes or FRED.

**Explain briefly about word2vec**

Word2Vec embeds words in a lower-dimensional vector space using a shallow neural network.

The result is a set of word-vectors where vectors close together in vector space have similar meanings based on context, and word-vectors distant to each other have differing meanings. For example, apple and orange would be close together and apple and gravity would be relatively far.

There are two versions of this model based on skip-grams (SG) and continuous-bag-of-words (CBOW).

**What are the metrics used to test an NLP model?**

Accuracy, Precision, Recall and F1. Accuracy is the usual ratio of the prediction to the desired output. But going just be accuracy is naive considering the complexities involved.

**What are some ways we can preprocess text input?**

Here are several preprocessing steps that are commonly used for NLP tasks:

● case normalization: we can convert all input to the same case (lowercase or uppercase) as a way of reducing our text to a more canonical form

● punctuation/stop word/white space/special characters removal: if we don’t think these words or characters are relevant, we can remove them to reduce the feature space

● lemmatizing/stemming: we can also reduce words to their inflectional forms (i.e. walks → walk) to further trim our vocabulary

● generalizing irrelevant information: we can replace all numbers with a <NUMBER> token or all names with a <NAME> token.

**How does the encoder-decoder structure work for language modelling?**

The encoder-decoder structure is a deep learning model architecture responsible for several state of the art solutions, including Machine Translation.

The input sequence is passed to the encoder where it is transformed to a fixed-dimensional vector representation using a neural network. The transformed input is then decoded using another neural network. Then, these outputs undergo another transformation and a SoftMax layer. The final output is a vector of probabilities over the vocabularies. Meaningful information is extracted based on these probabilities.

**How would you implement an NLP system as a service, and what are some pitfalls you might face in production?**

This is less of a NLP question than a question for productionizing machine learning models. There are however certain intricacies to NLP models.

Without diving too much into the productionization aspect, an ideal Machine Learning service will have:

● endpoint(s) that other business systems can use to make inference

● a feedback mechanism for validating model predictions

● a database to store predictions and ground truths from the feedback

● a workflow orchestrator which will (upon some signal) re-train and load the new model for

serving based on the records from the database + any prior training data

● some form of model version control to facilitate rollbacks in case of bad deployments

● post-production accuracy and error monitoring

**What are attention mechanisms and why do we use them?**

This was a follow-up to the encoder-decoder question. Only the output from the last time step is passed to the decoder, resulting in a loss of information learned at previous time steps. This information loss is compounded for longer text sequences with more time steps.

Attention mechanisms are a function of the hidden weights at each time step. When we use attention in encoder-decoder networks, the fixed-dimensional vector passed to the decoder becomes a function of all vectors outputted in the intermediary steps.

Two commonly used attention mechanisms are additive attention and multiplicative attention. As the names suggest, additive attention is a weighted sum while multiplicative attention is a weighted multiplier of the hidden weights. During the training process, the model also learns weights for the attention mechanisms to recognize the relative importance of each time step.

**How can we handle misspellings for text input?**

By using word embeddings trained over a large corpus (for instance, an extensive web scrape of billions of words), the model vocabulary would include common misspellings by design. The model can then learn the relationship between misspelled and correctly spelled words to recognize their semantic similarity.

We can also preprocess the input to prevent misspellings. Terms not found in the model vocabulary can be mapped to the “closest” vocabulary term using:

● edit distance between strings

● phonetic distance between word pronunciations

● keyword distance to catch common typos

**What is the problem with ReLu?**

● Exploding gradient(Solved by gradient clipping)

● Dying ReLu — No learning if the activation is 0 (Solved by parametric relu)

● Mean and variance of activations is not 0 and 1.(Partially solved by subtracting around 0.5 from activation. Better explained in fastai videos)

**What is the difference between learning latent features using SVD and getting embedding vectors using deep network?**

SVD uses linear combination of inputs while a neural network uses nonlinear combination.

**What is the information in the hidden and cell state of LSTM?**

Hidden stores all the information till that time step and cell state stores particular information that might be needed in the future time step.

**When is self-attention not faster than recurrent layers?**

When the sequence length is greater than the representation dimensions. This is rare.

**What is the benefit of learning rate warm-up?**

Learning rate warm-up is a learning rate schedule where you have low (or lower) learning rate at the beginning of training to avoid divergence due to unreliable gradients at the beginning. As the model becomes more stable, the learning rate would increase to speed up convergence.

**What’s the difference between hard and soft parameter sharing in multi-task learning?**

**What’s the difference between BatchNorm and LayerNorm?**

BatchNorm computes the mean and variance at each layer for every minibatch whereas LayerNorm computes the mean and variance for every sample for each layer independently.

Hard sharing is where we train for all the task at the same time and update our weights using all the losses whereas soft sharing is where we train for one task at a time.

Batch normalisation allows you to set higher learning rates, increasing speed of training as it reduces the unstability of initial starting weights.

**Difference between BatchNorm and LayerNorm?**

BatchNorm — Compute the mean and var at each layer for every minibatch

LayerNorm — Compute the mean and var for every single sample for each layer independently

**Why does the transformer block have LayerNorm instead of BatchNorm?**

Looking at the advantages of LayerNorm, it is robust to batch size and works better as it works at the sample level and not batch level.

**What changes would you make to your deep learning code if you knew there are errors in your training data?**

We can do label smoothening where the smoothening value is based on % error. If any particular class has known error, we can also use class weights to modify the loss.

What are the tricks used in ULMFiT? (Not a great questions but checks the awareness)

● LM tuning with task text

● Weight dropout

● Discriminative learning rates for layers

● Gradual unfreezing of layers

● Slanted triangular learning rate schedule

This can be followed up with a question on explaining how they help.

**Tell me a language model which doesn’t use dropout**

ALBERT v2 — This throws a light on the fact that a lot of assumptions we take for granted are not necessarily true. The regularization effect of parameter sharing in ALBERT is so strong that dropouts are not needed. (ALBERT v1 had dropouts.)

**What are the differences between GPT and GPT-2?**

● Layer normalization was moved to the input of each sub-block, similar to a residual unit of type “building block” (differently from the original type “bottleneck”, it has batch normalization applied before weight layers).

● An additional layer normalization was added after the final self-attention block.

● A modified initialization was constructed as a function of the model depth.

● The weights of residual layers were initially scaled by a factor of 1/√n where n is the number of residual layers.

● Use larger vocabulary size and context size.

**What are the differences between GPT and BERT?**

● GPT is not bidirectional and has no concept of masking

● BERT adds next sentence prediction task in training and so it also has a segment embedding

**What are the differences between BERT and ALBERT v2?**

● Embedding matrix factorisation(helps in reducing no. of parameters)

● No dropout

● Parameter sharing(helps in reducing no. of parameters and regularisation)

**How does parameter sharing in ALBERT affect the training and inference time?**

No effect. Parameter sharing just decreases the number of parameters.

**How would you reduce the inference time of a trained NN model?**

● Serve on GPU/TPU/FPGA

● 16 bit quantisation and served on GPU with fp16 support

● Pruning to reduce parameters

● Knowledge distillation (To a smaller transformer model or simple neural network)

● Hierarchical softmax/Adaptive softmax

● You can also cache results as explained here.

**Would you use BPE with classical models?**

Of course! BPE is a smart tokeniser and it can help us get a smaller vocabulary which can help us find a model with less parameters.

**How would you make an arxiv papers search engine? **

**How would you make a plagiarism detector?**

Get top k results with TF-IDF similarity and then rank results with

● semantic encoding + cosine similarity

● a model trained for ranking

**How would you make a sentiment classifier?**

This is a trick question. The interviewee can say all things such as using transfer learning and latest models but they need to talk about having a neutral class too otherwise you can have really good accuracy/f1 and still, the model will classify everything into positive or negative.

The truth is that a lot of news is neutral and so the training needs to have this class. The interviewee should also talk about how he will create a dataset and his training strategies like the selection of language model, language model fine-tuning and using various datasets for multitask learning.

**What is the difference between regular expression and regular grammar?**

A regular expression is the representation of natural language in the form of mathematical expressions containing a character sequence. On the other hand, regular grammar is the generator of natural language, defining a set of defined rules and syntax which the strings in the natural language must follow.

**Why should we use Batch Normalization?**

Once the interviewer has asked you about the fundamentals of deep learning architectures, they would move on to the key topic of improving your deep learning model’s performance.

Batch Normalization is one of the techniques used for reducing the training time of our deep learning algorithm. Just like normalizing our input helps improve our logistic regression model, we can normalize the activations of the hidden layers in our deep learning model as well:

**How is backpropagation different in RNN compared to ANN?**

In Recurrent Neural Networks, we have an additional loop at each node:

This loop essentially includes a time component into the network as well. This helps in capturing sequential information from the data, which could not be possible in a generic artificial neural network.

This is why the backpropagation in RNN is called Backpropagation through Time, as in backpropagation at each time step.

**Which of the following is a challenge when dealing with computer vision problems?**

Variations due to geometric changes (like pose, scale, etc), Variations due to photometric factors (like illumination, appearance, etc) and Image occlusion. All the above-mentioned options are challenges in computer vision.

**Consider an image with width and height as 100×100. Each pixel in the image can have a color from Grayscale, i.e. values. How much space would this image require for storing?**

The answer will be 8x100x100 because 8 bits will be required to represent a number from 0-256

**Why do we use convolutions for images rather than just FC layers?**

Firstly, convolutions preserve, encode, and actually use the spatial information from the image. If we used only FC layers we would have no relative spatial information. Secondly, Convolutional Neural Networks (CNNs) have a partially built-in translation in-variance, since each convolution kernel acts as it’s own filter/feature detector

**What makes CNN’s translation-invariant?**

As explained above, each convolution kernel acts as it’s own filter/feature detector. So let’s say you’re doing object detection, it doesn’t matter where in the image the object is since we’re going to apply the convolution in a sliding window fashion across the entire image anyways.

**Why do we have max-pooling in classification CNNs?**

Max-pooling in a CNN allows you to reduce computation since your feature maps are smaller after the pooling. You don’t lose too much semantic information since you’re taking the maximum activation. There’s also a theory that max-pooling contributes a bit to giving CNN’s more translation in-variance. Check out this great video from Andrew Ng on the benefits of max-pooling.

**Why do segmentation CNN’s typically have an encoder-decoder style/structure?**

The encoder CNN can basically be thought of as a feature extraction network, while the decoder uses that information to predict the image segments by “decoding” the features and upscaling to the original image size.

**What is the significance of Residual Networks?**

The main thing that residual connections did was allow for direct feature access from previous layers. This makes information propagation throughout the network much easier. One very interesting paper about this shows how using local skip connections gives the network a type of ensemble multi-path structure, giving features multiple paths to propagate throughout the network.

**What is batch normalization and why does it work?**

Training Deep Neural Networks is complicated by the fact that the distribution of each layer’s inputs changes during training, as the parameters of the previous layers change. The idea is then to normalize the inputs of each layer in such a way that they have a mean output activation of zero and a standard deviation of one. This is done for each individual mini-batch at each layer i.e compute the mean and variance of that mini-batch alone, then normalize. This is analogous to how the inputs to networks are standardized. How does this help? We know that normalizing the inputs to a network helps it learn.

But a network is just a series of layers, where the output of one layer becomes the input to the next. That means we can think of any layer in a neural network as the first layer of a smaller subsequent network. Thought of as a series of neural networks feeding into each other, we normalize the output of one layer before applying the activation function and then feed it into the following layer (sub-network).

**Why would you use many small convolutional kernels such as 3×3 rather than a few large ones?**

This is very well explained in the VGGNet paper.

There are 2 reasons: First, you can use several smaller kernels rather than few large ones to get the same receptive field and capture more spatial context, but with the smaller kernels you are using less parameters and computations. Secondly, because with smaller kernels you will be using more filters, you’ll be able to use more activation functions and thus have a more discriminative mapping function being learned by your CNN.

**What is Precision?**

Precision (also called positive predictive value) is the fraction of relevant instances among the retrieved instances

Precision = true positive / (true positive + false positive)

**What is Recall?**

Recall (also known as sensitivity) is the fraction of relevant instances that have been retrieved over the total amount of relevant instances.

Recall = true positive / (true positive + false negative)

**Define F1-score.**

It is the weighted average of precision and recall. It considers both false positive and false negatives into account. It is used to measure the model’s performance.

**What is cost function?**

The cost function is a scalar function that Quantifies the error factor of the Neural Network. Lower the cost function better than the Neural network. Eg: MNIST Data set to classify the image, the input image is digit 2 and the Neural network wrongly predicts it to be 3.

**List different activation neurons or functions**

● Linear Neuron

● Binary Threshold Neuron

● Stochastic Binary Neuron

● Sigmoid Neuron

● Tanh function

● Rectified Linear Unit (ReLU)

**Define Learning rate**

The learning rate is a hyper-parameter that controls how much we are adjusting the weights of our network with respect to the loss gradient.

**What is Momentum (w.r.t NN optimization)?**

Momentum lets the optimization algorithm remembers its last step, and adds some proportion of it to the current step. This way, even if the algorithm is stuck in a flat region, or a small local minimum, it can get out and continue towards the true minimum.

**What is the difference between Batch Gradient Descent and Stochastic Gradient Descent?**

Batch gradient descent computes the gradient using the whole dataset. This is great for convex or relatively smooth error manifolds. In this case, we move somewhat directly towards an optimum solution, either local or global. Additionally, batch gradient descent, given an annealed learning rate, will eventually find the minimum located in its basin of attraction.

Stochastic gradient descent (SGD) computes the gradient using a single sample. SGD works well (Not well, I suppose, but better than batch gradient descent) for error manifolds that have lots of local maxima/minima. In this case, the somewhat noisier gradient calculated using the reduced number of samples tends to jerk the model out of local minima into a region that hopefully is more optimal.

**Epoch vs Batch vs Iteration.**

Epoch: one forward pass and one backward pass of all the training examples

Batch: examples processed together in one pass (forward and backward)

Iteration: number of training examples / Batch size

**What is the vanishing gradient?**

As we add more and more hidden layers, backpropagation becomes less and less useful in passing information to the lower layers. In effect, as information is passed back, the gradients begin to vanish and become small relative to the weights of the networks.

**What are dropouts?**

Dropout is a simple way to prevent a neural network from overfitting. It is the dropping out of some of the units in a neural network. It is similar to the natural reproduction process, where nature produces offsprings by combining distinct genes (dropping out others) rather than strengthening the co-adapting of them.

**What is data augmentation? Can you give some examples?**

Data augmentation is a technique for synthesizing new data by modifying existing data in such a way that the target is not changed, or it is changed in a known way. Computer vision is one of the fields where data augmentation is very useful. There are many modifications that we can do to images:

● Resize

● Horizontal or vertical flip

● Rotate, Add noise, Deform

● Modify colors Each problem needs a customized data augmentation pipeline. For example, on OCR, doing flips will change the text and won’t be beneficial; however, resizes and small rotations may help.

**What are the components of GAN?**

● Generator

● Discriminator

**What’s the difference between a generative and discriminative model?**

A generative model will learn categories of data while a discriminative model will simply learn the distinction between different categories of data. Discriminative models will generally outperform generative models on classification tasks.

**What is Linear Filtering?**

Linear filtering is a neighborhood operation, which means that the output of a pixel’s value is decided by the weighted sum of the values of the input pixels.

**How can you achieve Blurring through Gaussian Filter?**

This is the most common technique for blurring or smoothing an image. This filter improves the resulting pixel found at the center and slowly minimizes the effects as pixels move away from the center. This filter can also help in removing noise in an image.

**How can you achieve Blurring through Gaussian Filter?**

This is the most common technique for blurring or smoothing an image. This filter improves the resulting pixel found at the center and slowly minimizes the effects as pixels move away from the center. This filter can also help in removing noise in an image.

**What is Non-Linear Filtering? How it is used?**

Linear filtering is easy to use and implement. In some cases, this method is enough to get the necessary output. However, an increase in performance can be obtained through non-linear filtering. Through non-linear filtering, we can have more control and achieve better results when we encounter a more complex computer vision task.

**Explain Median Filtering.**

The median filter is an example of a non-linear filtering technique. This technique is commonly used for minimizing the noise in an image. It operates by inspecting the image pixel by pixel and taking the place of each pixel’s value with the value of the neighboring pixel median.

Some techniques in detecting and matching features are:

● Lucas-Kanade

● Harris

● Shi-Tomasi

● SUSAN (smallest uni value segment assimilating nucleus)

● MSER (maximally stable extremal regions)

● SIFT (scale-invariant feature transform)

● HOG (histogram of oriented gradients)

● FAST (features from accelerated segment test)

● SURF (speeded-up robust features)

**Describe the Scale Invariant Feature Transform (SIFT) algorithm**

SIFT solves the problem of detecting the corners of an object even if it is scaled. Steps to implement this algorithm:

● Scale-space extrema detection – This step will identify the locations and scales that can still be recognized from different angles or views of the same object in an image.

● Keypoint localization – When possible key points are located, they would be refined to get accurate results. This would result in the elimination of points that are low in contrast or points that have edges that are deficiently localized.

● Orientation assignment – In this step, a consistent orientation is assigned to each key point to attain invariance when the image is being rotated.

● Keypoint matching – In this step, the key points between images are now linked to recognizing their nearest neighbors.

**Why Speeded-Up Robust Features (SURF) came into existence?**

SURF was introduced to as a speed-up version of SIFT. Though SIFT can detect and describe key points of an object in an image, still this algorithm is slow.

**What is Oriented FAST and rotated BRIEF (ORB)?**

This algorithm is a great possible substitute for SIFT and SURF, mainly because it performs better in computation and matching. It is a combination of fast key point detector and brief descriptor, which contains a lot of alterations to improve performance. It is also a great alternative in terms of cost because the SIFT and SURF algorithms are patented, which means that you need to buy them for their utilization.

**What is image segmentation?**

In computer vision, segmentation is the process of extracting pixels in an image that is related.

Segmentation algorithms usually take an image and produce a group of contours (the boundary of an object that has well-defined edges in an image) or a mask where a set of related pixels are assigned to a unique color value to identify it.

Popular image segmentation techniques:

● Active contours

● Level sets

● Graph-based merging

● Mean Shift

● Texture and intervening contour-based normalized cuts

**What is the purpose of semantic segmentation?**

The purpose of semantic segmentation is to categorize every pixel of an image to a certain class or label. In semantic segmentation, we can see what is the class of a pixel by simply looking directly at the color, but one downside of this is that we cannot identify if two colored masks belong to a certain object.

**Explain instance segmentation.**

In semantic segmentation, the only thing that matters to us is the class of each pixel. This would somehow lead to a problem that we cannot identify if that class belongs to the same object or not.

Semantic segmentation cannot identify if two objects in an image are separate entities. So to solve this problem, instance segmentation was created. This segmentation can identify two different objects of the same class. For example, if an image has two sheep in it, the sheep will be detected and masked with different colors to differentiate what instance of a class they belong to.

**How is panoptic segmentation different from semantic/instance segmentation?**

Panoptic segmentation is basically a union of semantic and instance segmentation. In panoptic segmentation, every pixel is classified by a certain class and those pixels that have several instances of a class are also determined. For example, if an image has two cars, these cars will be masked with different colors. These colors represent the same class — car — but point to different instances of a certain class.

**Explain the problem of recognition in computer vision.**

Recognition is one of the toughest challenges in the concepts in computer vision. Why is recognition hard? For the human eyes, recognizing an object’s features or attributes would be very easy. Humans can recognize multiple objects with very small effort. However, this does not apply to a machine. It would be very hard for a machine to recognize or detect an object because these objects vary. They vary in terms of viewpoints, sizes, or scales. Though these things are still challenges faced by most computer vision systems, they are still making advancements or approaches for solving these daunting tasks.

**What is Object Recognition?**

Object recognition is used for indicating an object in an image or video. This is a product of machine learning and deep learning algorithms. Object recognition tries to acquire this innate human ability, which is to understand certain features or visual detail of an image.

**What is Object Detection and it’s real-life use cases?**

Object detection in computer vision refers to the ability of machines to pinpoint the location of an object in an image or video. A lot of companies have been using object detection techniques in their system. They use it for face detection, web images, and security purposes.

**Describe Optical Flow, its uses, and assumptions.**

Optical flow is the pattern of apparent motion of image objects between two consecutive frames caused by the movement of object or camera. It is a 2D vector field where each vector is a displacement vector showing the movement of points from the first frame to the second

Optical flow has many applications in areas like :

● Structure from Motion

● Video Compression

● Video Stabilization

Optical flow works on several assumptions:

1. The pixel intensities of an object do not change between consecutive frames.

2. Neighboring pixels have similar motion.

**What is Histogram of Oriented Gradients (HOG)?**

HOG stands for Histograms of Oriented Gradients. HOG is a type of “feature descriptor”. The intent of a feature descriptor is to generalize the object in such a way that the same object (in this case a person) produces as close as possible to the same feature descriptor when viewed under different conditions. This makes the classification task easier.

**What’s the difference between valid and same padding in a CNN?**

This question has more chances of being a follow-up question to the previous one. Or if you have explained how you used CNNs in a computer vision task, the interviewer might ask this question along with the details of the padding parameters.

● Valid Padding: When we do not use any padding. The resultant matrix after convolution will have dimensions (n – f + 1) X (n – f + 1)

● Same padding: Adding padded elements all around the edges such that the output matrix will have the same dimensions as that of the input matrix

**What is BOV: Bag-of-visual-words (BOV)?**

BOV also called the bag of key points, is based on vector quantization. Similar to HOG features, BOV features are histograms that count the number of occurrences of certain patterns within a patch of the image.

**What is Poselets? Where are poselets used?**

Poselets rely on manually added extra keypoints such as “right shoulder”, “left shoulder”, “right knee” and “left knee”. They were originally used for human pose estimation

**Explain Textons in context of CNNs**

A texton is the minimal building block of vision. The computer vision literature does not give a strict definition for textons, but edge detectors could be one example. One might argue that deep learning techniques with Convolution Neuronal Networks (CNNs) learn textons in the first filters.

**What is Markov Random Fields (MRFs)?**

MRFs are undirected probabilistic graphical models which are a wide-spread model in computer vision. The overall idea of MRFs is to assign a random variable for each feature and a random variable for each pixel.

**Explain the concept of superpixel?**

A superpixel is an image patch that is better aligned with intensity edges than a rectangular patch.

Superpixels can be extracted with any segmentation algorithm, however, most of them produce highly irregular superpixels, with widely varying sizes and shapes. A more regular space tessellation may be desired.

**What is Non-maximum suppression(NMS) and where is it used?**

NMS is often used along with edge detection algorithms. The image is scanned along the image gradient direction, and if pixels are not part of the local maxima they are set to zero. It is widely used in object detection algorithms.

**Describe the use of Computer Vision in Healthcare.**

Computer vision has also been an important part of advances in health-tech. Computer vision algorithms can help automate tasks such as detecting cancerous moles in skin images or finding symptoms in x-ray and MRI scans

**Describe the use of Computer Vision in Augmented Reality & Mixed Reality**

Computer vision also plays an important role in augmented and mixed reality, the technology that enables computing devices such as smartphones, tablets, and smart glasses to overlay and embed virtual objects on real-world imagery. Using computer vision, AR gear detects objects in the real world in order to determine the locations on a device’s display to place a virtual object.

For instance, computer vision algorithms can help AR applications detect planes such as tabletops, walls, and floors, a very important part of establishing depth and dimensions and placing virtual objects in the physical world.

**Describe the use of Computer Vision in Facial Recognition**

Computer vision also plays an important role in facial recognition applications, the technology that enables computers to match images of people’s faces to their identities. Computer vision algorithms detect facial features in images and compare them with databases of face profiles.

Consumer devices use facial recognition to authenticate the identities of their owners. Social media apps use facial recognition to detect and tag users. Law enforcement agencies also rely on facial recognition technology to identify criminals in video feeds.

**Describe the use of Computer Vision in Self-Driving Cars**

Computer vision enables self-driving cars to make sense of their surroundings. Cameras capture video from different angles around the car and feed it to computer vision software, which then processes the images in real-time to find the extremities of roads, read traffic signs, detect other cars, objects, and pedestrians. The self-driving car can then steer its way on streets and highways, avoid hitting obstacles, and (hopefully) safely drive its passengers to their destination.

**Explain famous Computer Vision tasks using a single image example.**

Many popular computer vision applications involve trying to recognize things in photographs; for example:

Object Classification: What broad category of object is in this photograph?

Object Identification: Which type of a given object is in this photograph?

Object Verification: Is the object in the photograph?

Object Detection: Where are the objects in the photograph?

Object Landmark Detection: What are the key points for the object in the photograph?

Object Segmentation: What pixels belong to the object in the image?

Object Recognition: What objects are in this photograph and where are they?

**Explain the distinction between Computer Vision and Image Processing.**

Computer vision is distinct from image processing.

Image processing is the process of creating a new image from an existing image, typically simplifying or enhancing the content in some way. It is a type of digital signal processing and is not concerned with understanding the content of an image.

A given computer vision system may require image processing to be applied to raw input, e.g. pre-processing images.

Examples of image processing include:

● Normalizing photometric properties of the image, such as brightness or color.

● Cropping the bounds of the image, such as centering an object in a photograph.

● Removing digital noise from an image, such as digital artifacts from low light levels

**Explain business use cases in computer vision.**

● Optical character recognition (OCR)

● Machine inspection

● Retail (e.g. automated checkouts)

● 3D model building (photogrammetry)

● Medical imaging

● Automotive safety

● Match move (e.g. merging CGI with live actors in movies)

● Motion capture (mocap)

● Surveillance

● Fingerprint recognition and biometrics

**What is the Boltzmann Machine?**

One of the most basic Deep Learning models is a Boltzmann Machine, resembling a simplified version of the Multi-Layer Perceptron. This model features a visible input layer and a hidden layer — just a two-layer neural net that makes stochastic decisions as to whether a neuron should be on or off. Nodes are connected across layers, but no two nodes of the same layer are connected.

**What Is the Role of Activation Functions in a Neural Network?**

At the most basic level, an activation function decides whether a neuron should be fired or not. It accepts the weighted sum of the inputs and bias as input to any activation function. Step function,

Sigmoid, ReLU, Tanh, and Softmax are examples of activation functions.

**What Is the Difference Between a Feedforward Neural Network and Recurrent Neural Network?**

A Feedforward Neural Network signals travel in one direction from input to output. There are no feedback loops; the network considers only the current input. It cannot memorize previous inputs (e.g., CNN).

**What Are the Applications of a Recurrent Neural Network (RNN)?**

The RNN can be used for sentiment analysis, text mining, and image captioning. Recurrent Neural Networks can also address time series problems such as predicting the prices of stocks in a month or quarter.

**What Are the Softmax and ReLU Functions?**

Softmax is an activation function that generates the output between zero and one. It divides each output, such that the total sum of the outputs is equal to one. Softmax is often used for output layers.

**Machine Learning Techniques**

**What Is Overfitting, and How Can You Avoid It?**

Overfitting is a situation that occurs when a model learns the training set too well, taking up random fluctuations in the training data as concepts. These impact the model’s ability to generalize and don’t apply to new data.

When a model is given the training data, it shows 100 percent accuracy—technically a slight loss. But, when we use the test data, there may be an error and low efficiency. This condition is known as overfitting.

There are multiple ways of avoiding overfitting, such as:

● Regularization. It involves a cost term for the features involved with the objective function

● Making a simple model. With lesser variables and parameters, the variance can be reduced

● Cross-validation methods like k-folds can also be used

● If some model parameters are likely to cause overfitting, techniques for regularization like LASSO can be used that penalize these parameters

**What is meant by ‘Training set’ and ‘Test Set’?**

We split the given data set into two different sections namely, ‘Training set’ and ‘Test Set’.

‘Training set’ is the portion of the dataset used to train the model.

‘Testing set’ is the portion of the dataset used to test the trained model.

**How Do You Handle Missing or Corrupted Data in a Dataset?**

One of the easiest ways to handle missing or corrupted data is to drop those rows or columns or replace them entirely with some other value.

There are two useful methods in Pandas:

● IsNull() and dropna() will help to find the columns/rows with missing data and drop them

● Fillna() will replace the wrong values with a placeholder value

**How Do You Design an Email Spam Filter?**

Building a spam filter involves the following process:

● The email spam filter will be fed with thousands of emails

● Each of these emails already has a label: ‘spam’ or ‘not spam.’

● The supervised machine learning algorithm will then determine which type of emails are being marked as spam based on spam words like the lottery, free offer, no money, full refund, etc.

● The next time an email is about to hit your inbox, the spam filter will use statistical analysis and algorithms like Decision Trees and SVM to determine how likely the email is spam

● If the likelihood is high, it will label it as spam, and the email won’t hit your inbox

● Based on the accuracy of each model, we will use the algorithm with the highest accuracy after testing all the models

**Explain bagging.**

Bagging, or Bootstrap Aggregating, is an ensemble method in which the dataset is first divided into multiple subsets through resampling.

Then, each subset is used to train a model, and the final predictions are made through voting or averaging the component models.

Bagging is performed in parallel.

**What is the ROC Curve and what is AUC (a.k.a. AUROC)?**

The ROC (receiver operating characteristic) the performance plot for binary classifiers of True Positive Rate (y-axis) vs. False Positive Rate (xaxis).

AUC is the area under the ROC curve, and it’s a common performance metric for evaluating binary classification models.

It’s equivalent to the expected probability that a uniformly drawn random positive is ranked before a uniformly drawn random negative.

**What are the various Machine Learning algorithms?**

**What is cross-validation?**

Reference: k-fold cross validation

**Cross-validation is a resampling procedure used to evaluate machine learning models on a limited data sample.** The procedure has a single parameter called k that refers to the number of groups that a given data sample is to be split into. As such, the procedure is often called k-fold cross-validation. When a specific value for k is chosen, it may be used in place of k in the reference to the model, such as k=10 becoming 10-fold cross-validation. Mainly used in backgrounds where the objective is forecast, and one wants to estimate how accurately a model will accomplish in practice.

Cross-validation is primarily used in applied machine learning to estimate the skill of a machine learning model on unseen data. That is, to use a limited sample in order to estimate how the model is expected to perform in general when used to make predictions on data not used during the training of the model.

It is a popular method because it is simple to understand and because it generally results in a less biased or less optimistic estimate of the model skill than other methods, such as a simple train/test split.

The general procedure is as follows:

1. Shuffle the dataset randomly.

2. Split the dataset into k groups

3. For each unique group:

a. Take the group as a hold out or test data set

b. Take the remaining groups as a training data set

c. Fit a model on the training set and evaluate it on the test set

d. Retain the evaluation score and discard the model

4. Summarize the skill of the model using the sample of model evaluation scores

**What are 3 data preprocessing techniques to handle outliers?**

1. Winsorize (cap at threshold).

2. Transform to reduce skew (using Box-Cox or similar).

3. Remove outliers if you’re certain they are anomalies or measurement errors.

**How much data should you allocate for your training, validation, and test sets?**

You have to find a balance, and there’s no right answer for every problem.

If your test set is too small, you’ll have an unreliable estimation of model performance (performance statistic will have high variance). If your training set is too small, your actual model parameters will have a high variance.

A good rule of thumb is to use an 80/20 train/test split. Then, your train set can be further split into train/validation or into partitions for cross-validation.

**What Is a False Positive and False Negative and How Are They Significant?**

False positives are those cases which wrongly get classified as True but are False.

False negatives are those cases which wrongly get classified as False but are True.

In the term ‘False Positive’, the word ‘Positive’ refers to the ‘Yes’ row of the predicted value in

the confusion matrix. The complete term indicates that the system has predicted it as a positive, but the actual value is negative.

**What’s a Fourier transform?**

A Fourier transform is a generic method to decompose generic functions into a superposition of symmetric functions. Or as this more intuitive tutorial puts it, given a smoothie, it’s how we find the recipe. The Fourier transform finds the set of cycle speeds, amplitudes, and phases to match any time signal. A Fourier transform converts a signal from time to frequency domain — it’s a very common way to extract features from audio signals or other time series such as sensor data.

**Machine Learning Cheat Sheets, Tutorial, Practical examples, References, Datasets**

**Machine Learning Cheat Sheet**

Machine Learning Cheat Sheets**Credit: Remi Canard**

**TensorFlow Practical Examples and Tutorial**

– Basic Models

Linear Regression

Logistic Regression

Word2Vec (Word Embedding)

– Neural Networks

Simple Neural Network

Convolutional Neural Network

Recurrent Neural Network (LSTM)

Bi-directional Recurrent Neural Network (LSTM)

Dynamic Recurrent Neural Network (LSTM)

-Unsupervised

Auto-Encoder

DCGAN (Deep Convolutional Generative Adversarial Networks)

-Utilities:

Save and Restore a model

Build Custom Layers & Modules

– Data Management

Load and Parse data

Build and Load TFRecords

Image Transformation (i.e. Image Augmentation)

Credit: Alex Wang

**Cool MLOps repository of free talks, books, papers and more**

**Machine Learning Training Videos**

**References**

**1 https://springboard.com ****2 https://simplilearn.com ****3 https://geeksforgeeks.org ****4 https://elitedatascience.com ****5 https://analyticsvidhya.com ****6 https://guru99.com ****7 https://intellipaat.com ****8 https://towardsdatascience.com ****9 https://mygreatlearning.com****10 https://mindmajix.com ****11 https://toptal.com ****12 https://glassdoor.co.in ****13 https://udacity.com ****14 https://educba.com ****15 https://analyticsindiamag.com ****16 https://ubuntupit.com ****17 https://javatpoint.com ****18 https://quora.com ****19 hackr.io ****20 kaggle.com****21 https://www.linkedin.com/in/stevenouri/**

**Explain differences between AI, Machine Learning and NLP**

Artificial Intelligence |
Machine Learning |
Natural Language Processing |
---|---|---|

It is the technique t create smarter machines | Machine Learning is the term used for systems that learn from experience | This is the set of system that has the ability to understand the language |

AI includes human intervention | Machine Learning purely involves the working of computers and no human intervention | NLP links both computer and human languages |

Artificial intelligence is a broader concept than Machine Learning | ML is a narrow concept and is a subset of AI |

**Top Machine Learning Algorithms for Predictions:**

###
**Pick a smartphone for me: iPhone 12 Pro Max Reviews**
**Pick a smartphone for me: Xiaomi MI 10T Pro Reviews**
**Pick a smartphone for me: Sony Xperia 5 II Reviews**
**Pick a smartphone for me: Google Pixel 4a Reviews**
**Pick a smartphone for me: iPhone 12 Pro vs vs Galaxy Note Ultra 20 Reviews**
**Pick a smartphone for me: Google Pixel 5 XL Reviews**
**Pick a smartphone for me: Huawei P40 Pro Reviews**
**Pick a smartphone for me: Samsung Galaxy 20 Ultra Reviews**
**Motorola Edge 5G Reviews**

**Pick a smartphone for me: iPhone 12 Pro Max Reviews****Pick a smartphone for me: Xiaomi MI 10T Pro Reviews****Pick a smartphone for me: Sony Xperia 5 II Reviews****Pick a smartphone for me: Google Pixel 4a Reviews****Pick a smartphone for me: iPhone 12 Pro vs vs Galaxy Note Ultra 20 Reviews****Pick a smartphone for me: Google Pixel 5 XL Reviews****Pick a smartphone for me: Huawei P40 Pro Reviews****Pick a smartphone for me: Samsung Galaxy 20 Ultra Reviews****Motorola Edge 5G Reviews**