1-25

1. What is the primary use of 'gradient boosting' in machine learning?

o A. To remove noise from datasets

o B. To make a single model interpretable

o ✅ C. To combine multiple weak learners into a strong learner

o D. To increase feature importance

2. In ensemble learning, bagging and boosting are two different techniques.

What is the primary difference between them?
o ✅ A. Bagging combines multiple models to make predictions, while boosting
trains models sequentially.
o B. Bagging reduces bias, while boosting reduces variance.
o C. Boosting uses a single model, while bagging uses multiple models.
o D. Boosting is used only for regression, while bagging is for classification.

3. Which algorithm is commonly used for anomaly detection and one-class

classification? o A. K-Means Clustering

o B. Decision Trees

o ✅ C. One-Class Support Vector Machines (One-Class SVM)

o D. Logistic Regression

4. What is the purpose of dimensionality reduction techniques in machine

learning?

o A. To improve model interpretability by adding more parameters

o B. To increase the number of features

o ✅ C. To reduce the complexity of data while preserving important

information

o D. To remove all noise from data

5. In reinforcement learning, what is the term for the process of estimating

the expected cumulative rewards of taking a particular action in a specific
state?

o A. Exploration
o B. Temporal Difference Learning

o ✅ C. Q-Value

o D. Discount Factor

6. Which machine learning algorithm is known for its ability to handle

imbalanced datasets and is often used in fraud detection?

o A. K-Nearest Neighbors (KNN)

o B. Naive Bayes

o C. Logistic Regression

o ✅ D. Random Forest or Gradient Boosting

7. Which algorithm is commonly used for clustering and dimensionality

reduction and is based on finding the nearest neighbors of data points?

o ✅ A. DBSCAN (Density-Based Spatial Clustering of Applications with Noise)

o B. K-Means Clustering

o C. Principal Component Analysis (PCA)

o D. Decision Trees

8. What is the key idea behind word embeddings in natural language

processing?

o A. Representing words as one-hot encoded vectors

o B. Using only frequency-based word representations

o ✅ C. Representing words as dense vectors in a continuous space o

D. Sorting words alphabetically in datasets

9. What is the primary goal of 'convolutional neural networks' (CNNs) in deep

learning?

o A. To analyze tabular data

o B. To classify text documents

o ✅ C. To perform image recognition and feature extraction

o D. To process sequential data like time series

10. In reinforcement learning, what is the term for the numerical value that
represents the goodness or desirability of a state or action?

• ✅ A. Reward

• B. Discount Factor

• C. Policy Gradient

• D. Bellman

11. What is the primary goal of 'Kullback-Leibler divergence' (KL divergence)

in information theory?

• A. To find the shortest path between two data points

• ✅ B. To quantify the difference between two probability distributions

• C. To measure the variance of a dataset

• D. To optimize hyperparameters in a neural network

12. In deep learning, what is a convolutional layer primarily used for?

• ✅ A. Feature extraction from images

• B. Reducing noise in numerical data

• C. Combining multiple classifiers

• D. Splitting datasets into training and testing sets

13. What is the purpose of batch normalization in deep neural networks?

• A. To remove outliers from training data

• ✅ B. To help stabilize and accelerate training by normalizing input features

at each layer • C. To increase the number of neurons in each layer

• D. To reduce model accuracy for faster inference

14. In unsupervised learning, what is the main goal of 'hierarchical

clustering'?

• A. To assign each data point to a unique cluster

• B. To improve performance of supervised learning models

• ✅ C. To merge clusters based on similarity to create a hierarchy

• D. To train a deep learning model using small datasets

15. Which machine learning algorithm is often used for collaborative filtering
and recommendation systems?

• A. Decision Trees

• B. Logistic Regression

• ✅ C. Matrix Factorization (e.g., Singular Value Decomposition - SVD)

• D. K-Means Clustering

16. Which machine learning algorithm is often used for sentiment analysis
and text classification and is based on a probabilistic approach?

• A. Support Vector Machine (SVM)

• B. K-Nearest Neighbors (KNN)

• ✅ C. Naive Bayes

• D. Decision Trees

17. What is the main goal of semi-supervised learning?

• A. To train models using only labeled data

• ✅ B. To learn from both labeled and unlabeled data

• C. To reduce the number of features in a dataset

• D. To create synthetic data for training

18. What is the primary use of 'word embeddings' in natural language

processing?

• A. ✅ To represent words as dense vectors in a continuous space

• B. To convert words into one-hot encoded vectors

• C. To count the frequency of words in a document

• D. To group similar words using clustering

19. What is the primary objective of hyperparameter tuning in machine

learning?

• A. ✅ To improve model generalization

• B. To increase dataset size

• C. To remove bias from a dataset

• D. To train models faster without overfitting

20. What is the main benefit of using 'word2vec' embeddings in natural

language processing?

• A. They store words in tabular format

• ✅ B. They capture semantic relationships between words

• C. They reduce the need for labeled data

• D. They remove all redundant words from text

21. Which method is commonly used for handling missing data in a dataset?

• A. Deleting rows with missing values

• B. Replacing missing values with zero

• ✅ C. Imputation (e.g., mean, median, mode, interpolation)

• D. Ignoring missing values during training

22. In reinforcement learning, what is the term for the process of selecting
actions that maximize expected cumulative rewards while considering
exploration and exploitation?

• A. Reward Selection

• B. Gradient Descent

• C. Backpropagation

• ✅ D. Policy Optimization

23. What does the term 'overfitting' refer to in the context of machine
learning?

• A. When a model generalizes well to unseen data

• ✅ B. When a model performs well on training data but poorly on unseen

data

• C. When a model underestimates the complexity of a dataset

• D. When a model is unable to find a pattern in data

24. What is the primary objective of 'one-class classification' in machine

learning?

• A. To classify multiple categories of data

• B. ✅ To identify outliers and anomalies

• C. To increase training accuracy

• D. To reduce computation cost in training

25. What is the purpose of the 'sigmoid' activation function in neural

networks?

• A. ✅ To introduce non-linearity into the model

• B. To ensure weights do not change in a neural network

• C. To reduce overfitting in CNN models

• D. To increase training time

26. What is the primary goal of 'autoencoders' in deep learning?

• A. To classify images in supervised learning

• ✅ B. To compress data and reduce dimensionality

• C. To generate new labeled datasets

• D. To predict numerical values

27. Which machine learning technique is best suited for time series
forecasting and stock price prediction?

• A. Convolutional Neural Networks (CNNs)

• ✅ B. Recurrent Neural Networks (RNN) or Long Short-Term Memory (LSTM)

• C. Random Forest

• D. Naive Bayes

28. In reinforcement learning, what is the 'Bellman equation' used for?

• A. To increase dataset size

• ✅ B. To estimate the optimal value function

• C. To train deep neural networks

• D. To balance bias and variance

29. What is the primary use of a confusion matrix in classification tasks?

• A. To split data into training and testing sets

• ✅ B. To evaluate the performance of a classifier

• C. To increase model interpretability

• D. To visualize decision boundaries

30. What is the primary purpose of the 'mean squared error' (MSE) loss
function in regression tasks?

• A. ✅ To calculate the mean of squared errors between predicted and actual

values

• B. To improve model explainability

• C. To reduce bias in supervised learning

• D. To train deep learning models faster

31. Which deep learning architecture is designed to process sequences of

data and is commonly used in natural language processing?

• A. Convolutional Neural Networks (CNNs)

• ✅ B. Long Short-Term Memory (LSTM)

• C. Random Forest

• D. Naive Bayes

32. What is the primary use of 'online learning' in machine learning?

• A. To train models with static datasets

• ✅ B. To continuously update the model as new data becomes available

• C. To reduce model complexity

• D. To increase dataset interpretability

33. What is the primary advantage of using dropout layers in neural

networks?

• A. To speed up training time

• ✅ B. They prevent overfitting by randomly deactivating neurons during

training

• C. To reduce dataset size

• D. To replace backpropagation
34. In the context of reinforcement learning, what does the term 'exploration'
refer to?

• A. Avoiding new actions to reduce risk

• ✅ B. The process of selecting actions to discover new information

• C. Training a model using only supervised learning

• D. Increasing dataset complexity

35. What is the primary goal of unsupervised learning?

• A. To improve the performance of supervised models

• B. To train models using only labeled data

• ✅ C. To discover patterns and relationships in data

• D. To increase training speed in deep learning models

36. Which machine learning algorithm is commonly used for binary

classification problems and is based on a linear decision boundary?

• A. Decision Trees

• ✅ B. Support Vector Machine (SVM) or A. Logistic Regression

• C. K-Means Clustering

• D. Random Forest

37. What is the purpose of a 'one-hot encoding' in feature engineering?

• A. To compress high-dimensional data into a single value

• ✅ B. To convert categorical variables into a binary representation

• C. To cluster data points in an unsupervised manner

• D. To perform hyperparameter tuning

38. In natural language processing, what is the 'tf-idf' (term frequency-

inverse document frequency) measure used for?

• A. To count the number of times a word appears in a document

• ✅ B. To measure the importance of a word in a document relative to a

corpus

• C. To represent words as dense vectors

• D. To remove stopwords from text

39. What is the primary purpose of the 'softmax' activation function in neural
networks? • A. ✅ To convert logits into probabilities for multi-class
classification

• B. To normalize input data before training

• C. To reduce overfitting in deep learning models

• D. To balance the number of neurons in each layer

40. In reinforcement learning, what does the term 'policy' refer to?

• A. A dataset used for training

• B. A reinforcement learning algorithm

• ✅ C. A mapping from states to actions, defining the behavior of an agent

• D. A method for tuning hyperparameters

41. What is the primary challenge when training GANs?

• A. Lack of training data

• ✅ B. Mode collapse, where the generator produces limited diversity

• C. Slow training due to backpropagation

• D. High variance in predictions

42. Which machine learning model is commonly used for handling graph-
structured data? • A. ✅ Graph Neural Networks (GNNs)

• B. Decision Trees

• C. Random Forest

• D. Recurrent Neural Networks (RNNs)

43. What is the primary advantage of using Transformers over RNNs for NLP
tasks?

• A. They require fewer training examples

• B. They use convolutional layers instead of recurrent connections

• C. They process input sequences sequentially

• ✅ D. They can process the entire input sequence in parallel instead of
sequentially

44. Which method is commonly used for interpreting deep learning models?

• A. PCA (Principal Component Analysis)

• B. SHAP (Shapley Additive Explanations)

• ✅ C. LIME (Local Interpretable Model-agnostic Explanations)

• D. K-Means Clustering

45. What is the key advantage of Bayesian Optimization for hyperparameter

tuning?

• A. It reduces bias in models

• B. ✅ It efficiently searches for the best hyperparameters using probabilistic

models

• C. It requires minimal computational resources

• D. It is used only in supervised learning

46. In reinforcement learning, what is the main purpose of the 'reward

shaping' technique? • A. ✅ To accelerate learning by providing additional
guidance to the agent

• B. To increase exploration by penalizing incorrect actions

• C. To balance bias and variance in models

• D. To avoid overfitting in Q-learning

47. What is the main challenge of federated learning?

• A. Ensuring the dataset is large enough

• B. Reducing model overfitting

• ✅ C. Ensuring privacy and secure communication across distributed devices

• D. Increasing training speed

48. Which machine learning approach is best suited for multimodal learning?

• A. Using ensemble methods

• B. Training deep learning models with multiple architectures

• ✅ C. Combining multiple data sources such as text, images, and audio

• D. Using unsupervised learning exclusively

49. In explainable AI (XAI), what is the purpose of the LIME algorithm?

• A. To improve hyperparameter selection

• B. ✅ To approximate and explain model predictions using interpretable

models

• C. To detect and prevent bias in models

• D. To improve deep learning model generalization

50. What is the primary function of self-supervised learning?

• A. ✅ To learn representations from data without requiring labeled examples

• B. To train deep learning models using only supervised methods

• C. To ensure models are robust against adversarial attacks

• D. To classify images into pre-defined categories

51. What is a key application of contrastive learning in machine learning?

• A. To reduce the number of layers in deep networks

• B. To replace traditional loss functions in supervised learning

• ✅ C. Learning useful embeddings by pulling similar samples together and

pushing dissimilar ones apart

• D. To detect outliers in labeled datasets

52. Which optimization technique is best suited for training very deep neural
networks?

• A. Stochastic Gradient Descent (SGD)

• B. AdaGrad

• ✅ C. Adam optimizer

• D. K-Means

53. What is the primary role of zero-shot learning in NLP?

• A. To improve training speed

• ✅ B. To enable a model to classify data it has never seen before

• C. To replace labeled data with synthetic data

• D. To optimize the structure of neural networks

54. Which technique is most effective for handling catastrophic forgetting in

continual learning?

• A. Data Augmentation

• B. Hyperparameter Tuning

• ✅ C. Experience Replay

• D. Batch Normalization

55. What is the primary purpose of knowledge distillation in machine

learning?

• A. To increase dataset size

• B. To replace deep learning models with simpler models

• ✅ C. To transfer knowledge from a large model to a smaller model

• D. To remove redundant layers in deep networks

56. Which activation function is commonly used in transformers?

• A. ReLU

• B. Sigmoid

• ✅ C. GELU (Gaussian Error Linear Unit)

• D. Softmax

57. In reinforcement learning, what is 'multi-arm bandit' used for?

• A. Predicting time-series data

• B. Training generative adversarial networks

• ✅ C. Solving exploration-exploitation trade-offs in decision-making

• D. Hyperparameter tuning in neural networks

58. What is the primary advantage of using diffusion models in generative

AI?

• A. They require fewer labeled training samples

• B. They eliminate the need for adversarial training

• C. They process data faster than other generative models

• ✅ D. They generate high-quality images by modeling noise transitions over

time

59. Which type of neural network is used in AlphaFold for protein structure
prediction?

• A. Convolutional Neural Networks (CNNs)

• B. Recurrent Neural Networks (RNNs)

• ✅ C. Transformer-based models

• D. Generative Adversarial Networks (GANs)

60. What is the main role of masked language modeling in NLP?

• A. To reduce the number of tokens in a sentence

• B. ✅ To train models like BERT by predicting missing words in a sentence

• C. To segment text into predefined categories

• D. To convert text into one-hot encoded vectors

61. In adversarial machine learning, what is an 'evasion attack'?

• A. An attack that removes neurons from a network

• ✅ B. An attack that manipulates input data to fool a model at inference time

• C. An attack that reduces model training time

• D. An attack that modifies the loss function

62. Which deep learning framework is known for its symbolic differentiation
capability?

• A. TensorFlow

• ✅ B. JAX

• C. PyTorch

• D. Scikit-learn

63. What is the primary use of the Mixture of Experts (MoE) model in
machine learning?
• A. To increase model size for better accuracy

• B. To replace traditional neural networks

• ✅ C. To use multiple specialized subnetworks that activate based on input

data

• D. To fine-tune deep learning models

64. What is the advantage of meta-learning in few-shot learning tasks?

• A. ✅ It allows models to quickly adapt to new tasks with minimal data

• B. It increases dataset interpretability

• C. It reduces computation in deep learning models

• D. It replaces reinforcement learning algorithms

65. In unsupervised learning, what is the primary role of spectral clustering?

• A. To replace deep learning models in image classification

• B. To group labeled datasets efficiently

• ✅ C. Using eigenvalues of similarity matrices for clustering complex data

• D. To optimize the number of layers in a neural network

66. What is the key challenge of deploying deep learning models on edge
devices?

• A. Reducing dataset size

• B. ✅ Reducing computational and memory requirements

• C. Improving feature engineering

• D. Using more training samples

67. In reinforcement learning, what is 'curriculum learning'?

• A. A training method that ignores complex tasks

• ✅ B. A training strategy where tasks progressively increase in complexity

• C. A technique for reducing bias in machine learning models

• D. A method for feature selection

68. What is the purpose of an 'attention mechanism' in NLP?

• ✅ A. To focus on important parts of an input sequence while making
predictions

• B. To replace deep learning models in NLP

• C. To encode text into one-hot representations

• D. To generate word embeddings

69. Which metric is most suitable for evaluating generative models?

• A. Mean Squared Error (MSE)

• B. Accuracy

• C. ✅ Fréchet Inception Distance (FID) • D. Precision

70. What is the primary goal of entropy regularization in reinforcement

learning?

• A. To minimize training time

• B. To maximize accuracy

• ✅ C. Encouraging exploration by maximizing policy entropy

• D. To reduce computational complexity

71. What is the purpose of group normalization in deep learning?

• A. To increase model accuracy

• B. To reduce dataset imbalance

• ✅ C. To normalize activations by dividing features into groups instead of

using batch statistics

• D. To prevent adversarial attacks

72. Which framework is best suited for training physics-informed neural

networks (PINNs)? • A. PyTorch

• ✅ B. TensorFlow or JAX with differentiable physics solvers

• C. Scikit-learn

• D. XGBoost

73. What is the primary function of a hypernetwork in deep learning?

• A. To generate synthetic datasets

• ✅ B. To generate the weights of another neural network dynamically

• C. To replace the need for labeled data

• D. To improve dataset scaling

74. Which deep learning approach is best suited for learning hierarchical
structures in data?

• A. ✅ Capsule Networks

• B. Support Vector Machines (SVM)

• C. Random Forest

• D. Decision Trees

75. What is the main benefit of using the Lottery Ticket Hypothesis in deep
learning?

• A. ✅ It suggests that smaller sub-networks can be trained to achieve similar

performance as large networks

• B. It eliminates the need for labeled data

• C. It reduces training dataset size

• D. It ensures neural networks converge faster

76. In self-supervised learning, what is a common pretext task for learning

image representations?

• A. One-hot encoding of images

• B. Using decision trees for image classification

• C. Applying reinforcement learning on labeled images

• ✅ D. Predicting missing parts of an image or solving jigsaw puzzles

77. Which loss function is commonly used for training Variational

Autoencoders (VAEs)? • A. Cross-Entropy Loss

• ✅ B. Evidence Lower Bound (ELBO) loss

• C. Mean Squared Error (MSE)

• D. Hinge Loss

78. What is the main goal of entropy regularization in reinforcement

learning?
• A. To penalize incorrect actions

• ✅ B. Encouraging exploration by maximizing policy entropy

• C. To prevent underfitting

• D. To optimize hyperparameters

79. In multi-task learning, what is the main challenge of negative transfer?

• A. ✅ One task adversely affecting the learning of another task

• B. Reducing dataset size

• C. Training multiple tasks simultaneously

• D. Overfitting due to excessive data

80. What is a key advantage of using diffusion models for generative tasks?

• A. They eliminate noise from real images

• ✅ B. They generate high-quality samples by iteratively denoising random

noise

• C. They remove the need for training deep models

• D. They are less computationally expensive than GANs

81. Which method is commonly used for out-of-distribution detection?

• A. Cross-validation

• B. Gradient Boosting

• ✅ C. Mahalanobis distance

• D. Word2Vec

82. What is the primary purpose of quantization in deep learning?

• A. ✅ Reducing model size and computational cost

• B. Increasing accuracy of deep networks

• C. Improving image quality in generative models

• D. Eliminating adversarial attacks

83. Which framework is best suited for training physics-informed neural

networks (PINNs)? • A. PyTorch
• ✅ B. TensorFlow or JAX with differentiable physics solvers

• C. Scikit-learn

• D. XGBoost

84. What is the primary function of a hypernetwork in deep learning?

• A. To generate synthetic datasets

• ✅ B. To generate the weights of another neural network dynamically

• C. To replace the need for labeled data

• D. To improve dataset scaling

85. Which deep learning approach is best suited for learning hierarchical
structures in data?

• ✅ A. Capsule Networks

• B. Support Vector Machines (SVM)

• C. Random Forest

• D. Decision Trees

86. What is the purpose of curriculum learning in training deep learning

models?

• A. To randomly shuffle data to improve accuracy

• ✅ B. To progressively introduce harder tasks during training to improve

learning efficiency

• C. To create new labeled datasets

• D. To increase the number of parameters in a model

87. In few-shot learning, what is the purpose of prototypical networks?

• A. To perform reinforcement learning on small datasets

• B. To classify data using decision trees

• C. To optimize hyperparameters in deep learning models

• ✅ D. To classify new samples based on their proximity to learned class

prototypes
88. What is the primary goal of the Neural Tangent Kernel (NTK) in deep
learning?

• A. To train neural networks with fewer parameters

• ✅ B. To analyze the training dynamics of infinitely wide neural networks

• C. To optimize deep learning architectures for faster convergence

• D. To increase dataset complexity

89. In adversarial training, what is the primary purpose of the PGD (Projected
Gradient Descent) attack?

• ✅ A. To iteratively generate adversarial examples that maximize model

error

• B. To reduce overfitting in neural networks

• C. To improve the robustness of deep learning models

• D. To minimize loss functions in supervised learning

90. What is the main benefit of using the Lottery Ticket Hypothesis in deep
learning?

• A. ✅ It suggests that smaller sub-networks can be trained to achieve similar

performance as large networks

• B. It eliminates the need for labeled data

• C. It reduces training dataset size

• D. It ensures neural networks converge faster

91. Which type of neural network is most suitable for object detection in
images?

• A. Recurrent Neural Networks (RNNs)

• ✅ B. Convolutional Neural Networks (CNNs)

• C. Graph Neural Networks (GNNs)

• D. Long Short-Term Memory (LSTM) Networks

92. Which optimization method is best suited for reinforcement learning

tasks?

• A. Stochastic Gradient Descent (SGD)

• ✅ B. Trust Region Policy Optimization (TRPO)

• C. Principal Component Analysis (PCA)

• D. Support Vector Machines (SVM)

93. What is a key challenge in federated learning?

• A. Lack of labeled data

• ✅ B. Secure aggregation of distributed models while maintaining privacy

• C. Optimizing hyperparameters for deep learning models

• D. Generating adversarial examples

94. Which algorithm is commonly used for recommendation systems apart

from matrix factorization?

• A. Decision Trees

• B. Reinforcement Learning

• ✅ C. Deep Neural Networks (DNNs)

• D. K-Means Clustering

95. Which technique is commonly used to prevent mode collapse in GANs?

• A. Increasing the batch size

• ✅ B. Using Wasserstein GAN (WGAN) with gradient penalty

• C. Applying dropout layers

• D. Adding more convolutional layers

96. Which activation function is known for being computationally efficient in

deep networks?

• ✅ A. ReLU (Rectified Linear Unit)

• B. Sigmoid

• C. Tanh

• D. Softmax

97. Which reinforcement learning algorithm is model-based?

• A. Deep Q-Network (DQN)

• ✅ B. AlphaGo (Monte Carlo Tree Search + Deep Learning)

• C. Proximal Policy Optimization (PPO)

• D. Trust Region Policy Optimization (TRPO)

98. What is a primary advantage of meta-learning?

• A. It reduces the size of training datasets

• B. ✅ It enables fast adaptation to new tasks with minimal training examples

• C. It enhances feature selection in supervised learning

• D. It improves deep learning model accuracy

99. What is a key feature of contrastive divergence in machine learning?

• A. Used in reinforcement learning

• ✅ B. Used to efficiently train Restricted Boltzmann Machines (RBMs)

• C. Used to improve backpropagation in CNNs

• D. Used for dimensionality reduction

100. What is the primary advantage of using a Transformer over an LSTM for
NLP tasks?

• A. Faster training but higher computational cost

• ✅ B. Parallel processing of input sequences for faster computation

• C. Better suited for small datasets

• D. Less memory usage than LSTMs