AI & Machine Learning Interview Questions 2026

Answer: Supervised learning uses labeled data (examples with known answers) to train models. Unsupervised learning finds hidden patterns in unlabeled data. Supervised: spam detection, image classification. Unsupervised: customer segmentation, anomaly detection.

Answer: Overfitting occurs when a model performs well on training data but poorly on new data — it has memorized noise. Prevention techniques: regularization (L1/L2), dropout layers, cross-validation, early stopping, getting more training data.

Answer: Bias = error from overly simple models (underfitting). Variance = sensitivity to training data fluctuations (overfitting). The tradeoff: reducing bias often increases variance. A good model minimizes both for optimal generalization.

Answer: Precision: of all predicted positives, how many are actually positive. Recall: of all actual positives, how many did we predict correctly. F1 = harmonic mean of precision and recall — used when there is class imbalance.

Answer: A confusion matrix shows True Positives, True Negatives, False Positives, and False Negatives. It helps evaluate classification model performance beyond simple accuracy, especially for imbalanced datasets.

Answer: Hyperparameters are settings that control how a model learns (learning rate, number of layers, batch size). They are not learned from data. Tuning methods: Grid Search, Random Search, Bayesian Optimization.

Answer: Gradient descent is an optimization algorithm that minimizes the loss function by iteratively adjusting model weights in the direction of steepest descent (negative gradient). Variants: SGD, Mini-batch GD, Adam optimizer.

Answer: Transfer learning reuses a pre-trained model (e.g., ResNet, BERT) on a new task, rather than training from scratch. It is very useful when you have limited labeled data and dramatically reduces training time and cost.

Answer: CNN (Convolutional Neural Network) is designed for spatial data like images — it uses convolutional filters to detect local patterns. RNN (Recurrent Neural Network) handles sequential data like text and time series by maintaining a hidden state across time steps.

Answer: Regularization adds a penalty to the loss function to prevent overfitting. L1 regularization (Lasso) pushes some weights to zero for feature selection. L2 regularization (Ridge) reduces all weights proportionally. Dropout is regularization for neural networks.