Techniques

Supervised Learning

Training ML models on labelled input-output pairs to learn a function mapping inputs to target outputs.

Definition

Supervised learning is the most common ML paradigm: given a labelled training set {(x₁, y₁), ..., (xₙ, yₙ)}, learn a function f(x) → y that generalises to unseen inputs. Labels can be continuous (regression: predict house price) or categorical (classification: spam/not spam, image category).

The learned function is evaluated on a held-out test set to estimate generalisation performance. Overfitting (performing well on training but poorly on test) is controlled via regularisation (L1/L2 penalties, dropout), data augmentation, early stopping, and cross-validation.

Classic supervised algorithms include linear/logistic regression, decision trees, random forests, SVMs, and gradient boosted trees (XGBoost, LightGBM) — still widely used for tabular data. Deep learning is now dominant for perception tasks (images, text, audio) with enough data.

Examples

  • Email spam classifier
  • Image classification
  • Loan default prediction
  • Medical diagnosis from images