AI and machine learning interview questions

Supervised learning trains a model on labeled examples where each input already has a known target.

Unsupervised learning looks for structure in data that has no labels.

Reinforcement learning trains an agent to choose actions in an environment so that cumulative reward is maximized.

Classification predicts a discrete class such as spam or not spam.

Bias is error from assumptions that are too simple to capture the true pattern.

Variance is how much the fitted model jumps when the training sample changes.

Expected prediction error splits into bias, variance, and irreducible noise.

A straight line fit to a U-shaped curve is high bias.

Overfitting means the model has memorized training quirks instead of the general pattern.

Underfitting means the model is too simple or poorly trained to capture the signal.

Compare training metrics with validation metrics.

Collect more data, simplify the model, add L1 or L2 penalties, use dropout, prune trees, and stop training early.

The training set fits parameters.

Validation is used many times while you tune.

People often use about 70/15/15 or 80/10/10 when there is plenty of data.

Test information leaks into every modeling choice.

Accuracy looks strong on imbalanced data even if you always predict the majority class.

Precision is the share of predicted positives that are actually positive.

Recall is the share of actual positives that the model found.

F1 is the harmonic mean of precision and recall.