Machine Learning Questions
1. What is the primary goal of supervised learning?
A. To discover hidden patterns in unlabeled data
B. To learn a mapping from inputs to outputs using labeled examples
C. To optimize resource allocation in distributed systems
D. To generate synthetic data
B. To learn a mapping from inputs to outputs using labeled examples
2. Which learning paradigm uses rewards and penalties?
A. Supervised learning
B. Unsupervised learning
C. Reinforcement learning
D. Transfer learning
C. Reinforcement learning
3. High model complexity typically leads to:
A. Low bias and high variance
B. High bias and low variance
C. Low bias and low variance
D. High bias and high variance
A. Low bias and high variance
4. Which technique is NOT used to prevent overfitting?
A. Data augmentation
B. Increasing model parameters
C. Dropout
D. L2 regularization
B. Increasing model parameters
5. Which metric is most suitable for imbalanced classification?
A. Accuracy
B. F1-score
C. Mean squared error
D. R-squared
B. F1-score
6. The purpose of activation functions is to:
A. Increase computational speed
B. Introduce non-linearity
C. Reduce memory usage
D. Normalize input data
B. Introduce non-linearity
7. Stochastic gradient descent differs from batch gradient descent in:
A. Using the entire dataset for each update
B. Using a single random sample per update
C. Not requiring gradient computation
D. Being unsuitable for neural networks
B. Using a single random sample per update
8. The elbow method is used to:
A. Determine the optimal number of clusters
B. Initialize neural network weights
C. Select important features
D. Evaluate regression models
A. Determine the optimal number of clusters
9. PCA (Principal Component Analysis) works by:
A. Maximizing variance along orthogonal axes
B. Minimizing correlation between features
C. Removing categorical variables
D. Increasing feature dimensionality
A. Maximizing variance along orthogonal axes
10. Random Forests reduce variance by:
A. Using a single deep tree
B. Averaging predictions from many decorrelated trees
C. Eliminating all noisy features
D. Increasing model bias
B. Averaging predictions from many decorrelated trees
11. The kernel trick in SVM allows:
A. Linear separation in higher dimensions
B. Faster computation on GPUs
C. Elimination of support vectors
D. Reduction of training data size
A. Linear separation in higher dimensions
12. Pooling layers in CNNs primarily help with:
A. Increasing spatial resolution
B. Translation invariance and dimensionality reduction
C. Introducing non-linear activations
D. Generating new feature maps
B. Translation invariance and dimensionality reduction
13. The vanishing gradient problem affects:
A. Very deep feedforward networks only
B. Networks with ReLU activations only
C. RNNs processing long sequences
D. Decision trees
C. RNNs processing long sequences
14. L1 regularization (Lasso) tends to:
A. Produce dense weight vectors
B. Produce sparse weight vectors
C. Increase model variance
D. Be equivalent to L2 regularization
B. Produce sparse weight vectors
15. In AdaBoost, misclassified samples get:
A. Lower weights in subsequent iterations
B. Higher weights in subsequent iterations
C. Removed from the dataset
D. Randomly reassigned labels
B. Higher weights in subsequent iterations
16. Transfer learning is most beneficial when:
A. Target dataset is very large
B. Source and target domains are similar
C. Training time must be maximized
D. Models must be trained from scratch
B. Source and target domains are similar
17. Batch normalization helps by:
A. Reducing internal covariate shift
B. Increasing learning rate requirements
C. Slowing down training
D. Eliminating the need for activation functions
A. Reducing internal covariate shift
18. In Generative Adversarial Networks, the discriminator:
A. Only generates fake samples
B. Learns to distinguish real from fake samples
C. Is unnecessary after training
D. Uses reinforcement learning
B. Learns to distinguish real from fake samples
19. Which method is most efficient for hyperparameter tuning?
A. Grid search
B. Random search
C. Manual trial-and-error
D. Using default values only
B. Random search
20. Model fairness can be compromised by:
A. Using too much training data
B. Biases present in training data
C. High model accuracy
D. Small batch sizes
B. Biases present in training data