Question

Machine learning uses past examples to make predictions.

Answer 1

Yes — that’s the basic idea. Machine learning (ML) algorithms learn patterns from past data (the training set) and use those learned patterns to make predictions or decisions on new, unseen inputs.

Key points, briefly:
- Supervised learning: learns a mapping from inputs (features) to outputs (labels) using labeled past examples (e.g., image → “cat”).
- Unsupervised learning: finds structure in unlabeled data (e.g., clustering, dimensionality reduction).
- Reinforcement learning: learns policies from feedback (rewards) obtained by interacting with an environment.
- Training vs inference: training adjusts model parameters using past data; inference applies the trained model to make predictions.
- Important caveats: models must generalize to unseen data (not just memorize), can be biased if training data is biased, and can degrade if underlying patterns change (concept drift).

Examples: spam filters, recommendation systems, image classifiers, and language models all use past examples to predict future labels, scores, or actions.

Want a short example or a deeper dive into a specific ML type or concept?