Five ways a system can get better at a task from data or experience, instead of being handed an explicit rule for every case — each with its own home page, and its own place in a production AI stack.
A self-learning algorithm improves its own behaviour by processing data or experience, rather than following logic a programmer wrote by hand for every case. That single idea splits into five distinct paradigms below — and they aren't mutually exclusive. A deep neural network is a kind of machine learning model; several models can be combined into an ensemble; a reinforcement learning agent can itself be a deep network; and federated learning is a training arrangement that can apply to almost any of the others, training a shared model across many devices or institutions without centralizing their data.
Each card links to its own home page on peterindia.net, curating platforms, frameworks, algorithms, and resources for that paradigm specifically.
The broad discipline of models that learn statistical patterns from data — the parent category that deep learning and ensemble methods both sit inside.
Machine learning with multi-layered neural networks, which learn their own feature representations directly from raw data instead of needing them hand-engineered.
An agent learns by acting, observing, and being rewarded — no labelled dataset required. The paradigm behind game-playing agents, robotics control, and the RLHF pipelines that align large language models.
Combining many individually weaker models — through bagging, boosting, or stacking — into a single predictor that generalizes better than any one model alone.
A privacy-preserving training arrangement where a shared model learns across distributed devices or institutions without the raw data ever leaving its source — only model updates are exchanged and aggregated.