Machine Learning flashcards, generated for you.

A system that learns patterns from data and uses those patterns to make predictions or decisions on new, unseen data—without being explicitly programmed for each case.

Training data teaches the model patterns; testing data checks if it learned generalizable patterns (not just memorized training examples). This prevents overfitting.

Supervised learning uses labeled data (input→correct output) to learn a mapping. Unsupervised learning finds hidden patterns in unlabeled data without knowing the 'right answer.'

Overfitting occurs when a model memorizes noise and peculiarities in training data instead of learning general patterns. It performs well on training data but poorly on new data.

Bias = tendency to underfit (oversimplified model misses patterns). Variance = tendency to overfit (complex model captures noise). You must balance: too simple fails, too complex generalizes poorly.

A loss function quantifies how far a model's predictions are from the true values. Lower loss = better predictions. Minimizing loss is the goal of training.

An iterative optimization algorithm that adjusts model parameters in small steps in the direction that most reduces loss. It's efficient for finding good parameter values without checking every possibility.

Accuracy: (correct predictions)/(total predictions). Precision: (true positives)/(predicted positives)—how many predicted positive are actually positive. Recall: (true positives)/(actual positives)—how many actual positives did we find.

Regularization adds a penalty term to the loss function for complex models (e.g., large weights). This discourages the model from fitting noise and forces it toward simpler, more generalizable solutions.

Creating or transforming input variables (features) to make patterns easier for the model to learn. Good features are more predictive; the model learns what matters most from well-engineered data.