BSCS Bachelor of Science in Computer Science: Artificial Intelligence Fundamentals 3 — Questions and Answers
Question 1: In support vector machines (SVM), what is the role of the kernel function?
- To initialize weights before training
- To map data into a higher-dimensional feature space where it becomes linearly separable (Correct answer)
- To regularize the margin by penalizing support vectors
- To select the learning rate during gradient descent
Correct answer: To map data into a higher-dimensional feature space where it becomes linearly separable
Kernel functions implicitly compute dot products in a transformed feature space without explicitly mapping data, enabling non-linear classification.
Question 2: Which natural language processing technique represents words as dense vectors in a continuous space such that semantically similar words are geometrically close?
- TF-IDF weighting
- Bag-of-words encoding
- Word embeddings (e.g., Word2Vec) (Correct answer)
- One-hot encoding
Correct answer: Word embeddings (e.g., Word2Vec)
Word embeddings like Word2Vec learn dense vector representations where semantic relationships are captured by vector geometry.
Question 3: What does the 'exploration vs. exploitation' dilemma refer to in reinforcement learning?
- Whether to use model-based or model-free learning
- Balancing trying new actions to gather information versus using known high-reward actions (Correct answer)
- Deciding between on-policy and off-policy methods
- Choosing between discrete and continuous action spaces
Correct answer: Balancing trying new actions to gather information versus using known high-reward actions
An agent must balance exploiting known good actions for reward versus exploring unknown actions that might yield higher long-term reward.
Question 4: In a Hidden Markov Model (HMM), what are the 'hidden' elements?
- The observation symbols emitted at each time step
- The underlying state sequence that generates the observations (Correct answer)
- The transition probabilities between states
- The emission probabilities of each symbol
Correct answer: The underlying state sequence that generates the observations
HMMs have latent states that are not directly observable; only the emitted observations are visible to the learner.
Question 5: Which ensemble learning method trains multiple models sequentially, where each new model focuses on correcting errors made by previous models?
- Bagging
- Random Forest
- Boosting (Correct answer)
- Stacking
Correct answer: Boosting
Boosting (e.g., AdaBoost, Gradient Boosting) trains learners sequentially, with each one upweighting misclassified examples from prior models.
Question 6: What is the vanishing gradient problem in deep neural networks?
- Gradients explode to infinity during backpropagation in deep networks
- Gradients become extremely small during backpropagation, preventing early layers from learning (Correct answer)
- The loss function fails to converge to a local minimum
- Weight initialization causes outputs to saturate at layer boundaries
Correct answer: Gradients become extremely small during backpropagation, preventing early layers from learning
In deep networks, gradients are multiplied through many layers; with small derivatives (e.g., sigmoid), they shrink exponentially toward the input layers.
Question 7: Which constraint satisfaction technique eliminates values from domains that cannot participate in any valid solution before or during search?
- Forward checking
- Min-conflicts heuristic
- Arc consistency (AC-3) (Correct answer)
- Backjumping
Correct answer: Arc consistency (AC-3)
Arc consistency (AC-3) enforces constraints by removing domain values that have no compatible value in a neighboring variable's domain.
BSCS Bachelor of Science in Computer Science: Artificial Intelligence Fundamentals 3
In support vector machines (SVM), what is the role of the kernel function?