Machine Learning
Multiple Choice Questions

A subset of AI that enables systems to learn from data

A programming language for AI

A database management system

A type of hardware

Arthur Samuel

Alan Turing

Geoffrey Hinton

Andrew Ng

Supervised, Unsupervised, Reinforcement

Linear, Logistic, Neural

Classification, Regression, Clustering

Training, Testing, Validation

AI is the broader concept, ML is a subset of AI

ML is the broader concept, AI is a subset

They are the same thing

AI is older than ML

To measure how well the model performs

To optimize the model's parameters

To split the data

To visualize data

Model performs well on training data but poorly on new data

Model performs poorly on both training and new data

Model performs equally well on all data

Model fails to learn from training data

Learning from labeled data

Learning from unlabeled data

Learning through trial and error

Learning without any data

Logistic Regression

Linear Regression

K-Means

PCA

Classification predicts discrete labels, regression predicts continuous values

Regression predicts discrete labels, classification predicts continuous values

Both predict continuous values

Both predict discrete labels

A non-parametric algorithm that classifies based on nearest neighbors

A parametric algorithm that uses probability

A reinforcement learning algorithm

A deep learning algorithm

To evaluate model performance on unseen data

To increase training data

To reduce overfitting

To normalize data

A table used to evaluate classification model performance

A matrix for regression analysis

A visualization tool for clustering

A neural network architecture

Learning from unlabeled data

Learning from labeled data

Learning through trial and error

Learning with human feedback

K-Means

Linear Regression

Support Vector Machine

Random Forest

Reducing the number of features/variables

Reducing the number of data points

Reducing the number of classes

Reducing the model complexity

To reduce dimensionality and identify important features

To classify data

To cluster data

To predict continuous values

K-Means requires number of clusters, Hierarchical creates a hierarchy

Hierarchical requires number of clusters, K-Means creates a hierarchy

Both create hierarchies

Both require number of clusters

Identifying rare or unusual data points

Identifying common data points

Predicting future data points

Creating new data points

A method to model the relationship between variables using a linear equation

A classification algorithm

A clustering algorithm

A deep learning model

Mean Squared Error (MSE)

Cross-entropy loss

Hinge loss

Log loss

To minimize the cost function by updating parameters

To maximize the cost function

To split the data

To visualize the data

Simple uses one independent variable, multiple uses several

Multiple uses one independent variable, simple uses several

Both use one variable

Both use multiple variables

A measure of how well the model explains the variance

The error rate of the model

The number of features

The learning rate

Techniques to prevent overfitting by adding a penalty term

Techniques to increase overfitting

Techniques to reduce data size

Techniques to increase data size

A classification algorithm that uses a sigmoid function

A regression algorithm for continuous values

A clustering algorithm

A neural network

To map output to probabilities between 0 and 1

To map output to continuous values

To cluster data

To reduce dimensions

Logistic is for classification, linear is for regression

Linear is for classification, logistic is for regression

Both are for classification

Both are for regression

A classification algorithm that finds the optimal hyperplane

A clustering algorithm

A regression algorithm

A neural network

A method to transform data to higher dimensions without explicit computation

A method to reduce dimensions

A method to cluster data

A method to normalize data

Bagging trains models in parallel, boosting trains sequentially

Boosting trains models in parallel, bagging trains sequentially

Both train in parallel

Both train sequentially

A tree-like model that makes decisions based on feature values

A neural network architecture

A clustering algorithm

A regression algorithm

A measure of randomness or impurity

A measure of accuracy

A measure of precision

A measure of recall

The reduction in entropy after splitting on a feature

The increase in accuracy

The reduction in training time

The increase in data size

A measure of how often a randomly chosen element would be incorrectly classified

A measure of model accuracy

A measure of model precision

A measure of model recall

Random forest is an ensemble of multiple decision trees

Decision tree is an ensemble of random forests

Both are the same

Decision tree is more accurate

Removing branches to reduce overfitting

Adding branches to increase complexity

Increasing tree depth

Decreasing tree depth

To combine multiple models for better performance

To use a single model

To reduce data size

To increase data size

An ensemble of decision trees using bagging

A single decision tree

A neural network

A clustering algorithm

Bagging reduces variance, boosting reduces bias

Boosting reduces variance, bagging reduces bias

Both reduce variance

Both reduce bias

An optimized gradient boosting algorithm

A random forest implementation

A neural network library

A clustering algorithm

Adaptive Boosting that gives more weight to misclassified samples

A random forest algorithm

A neural network

A clustering algorithm

Combining multiple models using a meta-model

Training models in sequence

Training models in parallel

Using a single model

The ratio of correct predictions to total predictions

The ratio of false positives to true positives

The ratio of true positives to false positives

The ratio of false negatives to true negatives

The ratio of true positives to all positive predictions

The ratio of true positives to all actual positives

The ratio of correct predictions to all predictions

The ratio of false positives to true positives

The ratio of true positives to all actual positives

The ratio of true positives to all positive predictions

The ratio of correct predictions to all predictions

The ratio of false positives to true positives

The harmonic mean of precision and recall

The arithmetic mean of precision and recall

The geometric mean of precision and recall

The sum of precision and recall

A curve that plots true positive rate vs false positive rate

A curve that plots precision vs recall

A curve that plots accuracy vs error rate

A curve that plots loss vs epochs

A technique to evaluate model performance by splitting data into folds

A technique to increase data size

A technique to reduce features

A technique to optimize hyperparameters

A computational model inspired by the human brain

A linear regression model

A decision tree

A clustering algorithm

The simplest type of neural network with one layer

A deep neural network

A convolutional neural network

A recurrent neural network

To introduce non-linearity

To introduce linearity

To reduce dimensions

To cluster data

An algorithm to update weights by propagating error backwards

An algorithm to propagate input forward

An algorithm to reduce dimensions

An algorithm to cluster data

Shallow has few layers, deep has many layers

Deep has few layers, shallow has many layers

Both have the same number of layers

Shallow is more accurate

To control the step size in gradient descent

To control the number of epochs

To control the batch size

To control the number of layers

A subset of ML using deep neural networks

A subset of AI without neural networks

A type of clustering

A type of regression

A neural network designed for image and spatial data

A neural network for sequential data

A neural network for text data

A neural network for tabular data

A neural network designed for sequential data

A neural network for image data

A neural network for tabular data

A neural network with no connections

Gradients become too small for effective learning in deep networks

Gradients become too large

Gradients become zero

Gradients become infinite

To prevent overfitting by randomly dropping neurons

To increase model complexity

To reduce training time

To increase accuracy

Using knowledge from one task to improve performance on another

Learning from scratch

Learning without data

Learning with no labels

Mean is sensitive to outliers, median is robust

Median is sensitive to outliers, mean is robust

Both are sensitive to outliers

Both are robust to outliers

A measure of the spread of data around the mean

A measure of central tendency

A measure of correlation

A measure of probability

Correlation doesn't imply causation

Correlation implies causation

They are the same

Causation doesn't imply correlation

The probability of observing results as extreme as the observed, assuming null hypothesis is true

The probability that the null hypothesis is true

The probability of a Type I error

The probability of a Type II error

A symmetric bell-shaped distribution

A skewed distribution

A uniform distribution

A bimodal distribution

To calculate conditional probabilities for classification

To calculate means and variances

To calculate correlations

To calculate standard deviations

Normalizing features to a common range

Adding new features

Removing features

Creating features

Normalization scales to [0,1], standardization has mean 0, std 1

Standardization scales to [0,1], normalization has mean 0, std 1

Both are the same

Neither is used

Converting categorical variables to binary vectors

Converting continuous variables to categorical

Converting text to numbers

Converting numbers to text

Choosing the most relevant features for the model

Creating new features

Scaling features

Encoding features

To improve model performance by creating better features

To reduce data size

To increase data size

To remove all features

The problem of high-dimensional data leading to sparsity and overfitting

The problem of too little data

The problem of too many samples

The problem of missing values

The interaction between computers and human language

Processing numerical data

Processing images

Processing audio

Splitting text into smaller units like words or sentences

Converting text to numbers

Removing stop words

Lemmatizing words

Stemming is rule-based, lemmatization uses vocabulary and morphology

Lemmatization is rule-based, stemming uses vocabulary

Both are the same

Neither is used

A dense vector representation of words

A one-hot encoding of words

A bag-of-words representation

A TF-IDF representation

To remove common words that don't add significant meaning

To remove all words

To keep only stop words

To remove all content words

A neural network architecture using attention mechanisms

A traditional RNN architecture

A CNN architecture

A decision tree architecture

The field of AI that deals with visual data

The field of AI that deals with text

The field of AI that deals with audio

The field of AI that deals with tabular data

Assigning a class label to an image

Detecting objects in an image

Segmenting an image

Generating an image

Identifying and locating objects in an image

Classifying the entire image

Segmenting the image

Generating the image

Dividing an image into segments based on pixel groups

Classifying the entire image

Detecting objects in the image

Generating the image

To reduce spatial dimensions while preserving important features

To increase spatial dimensions

To add non-linearity

To normalize features

Detection gives bounding boxes, segmentation gives pixel-level masks

Segmentation gives bounding boxes, detection gives pixel-level masks

Both give bounding boxes

Both give pixel-level masks

Learning through interaction with an environment to maximize rewards

Learning from labeled data

Learning from unlabeled data

Learning without any data

The learner that interacts with the environment

The environment itself

The reward function

The policy

A strategy that the agent follows to decide actions

The reward function

The environment

The state space

Balancing between trying new actions and using known good actions

Balancing between training and testing

Balancing between supervised and unsupervised learning

Balancing between classification and regression

A value-based reinforcement learning algorithm

A policy-based reinforcement learning algorithm

A supervised learning algorithm

An unsupervised learning algorithm

Feedback from the environment indicating the quality of an action

The prediction error

The classification accuracy

The clustering quality

The practice of managing ML model lifecycle in production

The practice of training ML models

The practice of data engineering

The practice of model evaluation

Tracking different versions of ML models

Tracking data versions

Tracking code versions

Tracking hardware versions

The degradation of model performance over time

The improvement of model performance over time

The change in model architecture

The change in model parameters

To compare two models and choose the better one

To test model accuracy

To test model precision

To test model recall

Continuously tracking model performance and health

Training the model

Deploying the model

Testing the model

Training learns patterns, inference applies learned patterns to new data

Inference learns patterns, training applies learned patterns

Both learn patterns

Both apply learned patterns

Grouping similar data points together

Classifying data points into predefined categories

Predicting continuous values

Reducing data dimensions

A method to find the optimal number of clusters

A method to initialize centroids

A method to evaluate cluster quality

A method to visualize clusters

A measure of how similar a point is to its own cluster compared to others

A measure of cluster size

A measure of cluster density

A measure of cluster distance

K-Means is centroid-based, DBSCAN is density-based

DBSCAN is centroid-based, K-Means is density-based

Both are centroid-based

Both are density-based

Creating a hierarchy of clusters in a tree-like structure

Creating clusters of equal size

Creating clusters based on density

Creating clusters based on centroids

To visualize the hierarchy of clusters

To measure cluster quality

To initialize centroids

To assign cluster labels

Training data is used to learn, testing data is used to evaluate

Testing data is used to learn, training data is used to evaluate

Both are used for learning

Both are used for evaluation

The data points closest to the decision boundary in SVM

The data points farthest from the decision boundary

The centroids of clusters

The means of distributions

To allow the model to fit data that doesn't pass through the origin

To increase model complexity

To reduce overfitting

To increase accuracy

Binary has two classes, multi-class has more than two

Multi-class has two classes, binary has more than two

Both have two classes

Both have more than two classes

ID3 uses information gain, C4.5 uses gain ratio

C4.5 uses information gain, ID3 uses gain ratio

Both use information gain

Both use gain ratio

To perform slightly better than random guessing

To perform perfectly

To perform poorly

To perform randomly

Balanced accuracy handles class imbalance better

Accuracy handles class imbalance better

Both are the same

Neither handles class imbalance

To normalize inputs of each layer to improve training

To reduce model complexity

To increase model complexity

To reduce training time

CNN is for spatial data, RNN is for sequential data

RNN is for spatial data, CNN is for sequential data

Both are for spatial data

Both are for sequential data

Correlation is a normalized version of covariance

Covariance is a normalized version of correlation

Both are the same

Neither is normalized

To create non-linear features from linear ones

To reduce features

To encode categorical features

To scale features

To weight words based on importance in a document

To convert text to word embeddings

To remove stop words

To perform tokenization

To increase the diversity of training data

To reduce the size of training data

To remove noise from images

To compress images

On-policy uses current policy, off-policy can use different policies

Off-policy uses current policy, on-policy can use different policies

Both use current policy

Both use different policies

To automate model training and deployment

To manually train models

To manually deploy models

To evaluate models

Hard assigns each point to one cluster, soft assigns probabilities

Soft assigns each point to one cluster, hard assigns probabilities

Both assign to one cluster

Both assign probabilities

Parameters are learned, hyperparameters are set before training

Hyperparameters are learned, parameters are set before training

Both are learned

Both are set before training

To allow the model to fit data that doesn't pass through the origin

To increase model complexity

To reduce overfitting

To increase accuracy