Difficult Machine Learning MCQs

Which assumption is not true for the Naïve Bayes classifier?

• Features are independent of each other

• The dataset is balanced

• The probability distribution follows a Gaussian distribution

• All features have the same importance

Which of the following methods can be used to handle imbalanced datasets?

• SMOTE (Synthetic Minority Over-sampling Technique)

• Dropout Regularization

• Increasing batch size

• Weight decay

In a deep neural network, vanishing gradients are more common with which activation function?

• ReLU

• Sigmoid

• Leaky ReLU

• Softmax

Which of the following optimization techniques adapts the learning rate for each parameter separately?

• Adam

• Momentum

• Stochastic Gradient Descent (SGD)

• Newton’s Method

Which statement about the bias-variance tradeoff is incorrect?

• A high-bias model underfits the data

• A high-variance model overfits the data

• Increasing model complexity always reduces bias

• Reducing variance can sometimes increase bias

In a convolutional neural network (CNN), what is the purpose of pooling layers?

• To increase computation cost

• To extract low-level features

• To reduce spatial dimensions

• To perform backpropagation efficiently

Which of the following architectures is commonly used for sequence-to-sequence learning?

• CNN

• RNN

• GAN

• Autoencoder

Which technique prevents exploding gradients in deep networks?

• Batch Normalization

• Xavier Initialization

• Gradient Clipping

• Dropout

Which of the following is true about transformer-based models like BERT?

• They use RNNs for sequential processing

• They rely entirely on self-attention mechanisms

• They perform worse than LSTMs for large datasets

• They cannot be used for language translation

Which loss function is most commonly used for multi-class classification problems?

• Mean Squared Error

• Cross-Entropy Loss

• Hinge Loss

• L1 Loss