AI Interview Questions - Artificial Intelligence Interview Prep

Artificial Intelligence is broad field of creating intelligent machines that can perform tasks requiring human intelligence. Machine Learning is subset of AI that enables systems to learn from data without explicit programming. AI includes ML, but also rule-based systems, expert systems, and other approaches.

Supervised learning uses labeled data to train models that predict outputs. Unsupervised learning finds patterns in unlabeled data without target outputs. Supervised: classification, regression. Unsupervised: clustering, dimensionality reduction, association.

Overfitting occurs when model learns training data too well, including noise, and performs poorly on new data. Prevent with: cross-validation, regularization (L1/L2), dropout, early stopping, more data, feature selection, ensemble methods, reducing model complexity.

Cross-validation splits data into k folds, trains on k-1 folds, tests on remaining fold, repeats k times. Provides better estimate of model performance than single train/test split. Common: k-fold (k=5 or 10), stratified k-fold, leave-one-out.

Precision measures accuracy of positive predictions (TP / (TP + FP)). Recall measures ability to find all positives (TP / (TP + FN)). High precision: few false positives. High recall: few false negatives. F1-score balances both: 2 * (precision * recall) / (precision + recall).

Gradient descent is optimization algorithm that minimizes cost function by iteratively moving in direction of steepest descent (negative gradient). Updates parameters: θ = θ - α * ∇J(θ). α is learning rate. Variants: batch, stochastic, mini-batch, Adam, RMSprop.

Neural network is computing system inspired by biological neurons. Consists of layers: input, hidden, output. Each neuron applies activation function to weighted sum of inputs. Learns by adjusting weights through backpropagation. Can approximate any function (universal approximation theorem).

Backpropagation is algorithm for training neural networks. Calculates gradient of loss function with respect to weights using chain rule, propagating errors backward from output to input. Enables efficient computation of gradients for all layers. Core of deep learning training.

Activation functions introduce non-linearity to neural networks. Without them, network is just linear transformation. Common: ReLU (most popular), sigmoid (0-1), tanh (-1 to 1), softmax (probabilities). ReLU avoids vanishing gradient, faster training.

Batch uses all training data per update, stable but slow. Stochastic uses one sample per update, fast but noisy. Mini-batch uses small subset (32-256 samples), balances speed and stability. Mini-batch is most common in practice.