Skip to content

2. Machine Learning / ML

Systems that improve at a task from data rather than explicit programming. The children partition by supervision signal — where the learning signal comes from: labeled (supervised), none (unsupervised), self-generated (self-supervised), partial (semi-supervised), or reward (reinforcement). Reinforcement learning is special because it learns from interaction over time, hence its own sub-vocabulary (agent / environment / state / action / reward / policy), which reappears generalized in Agents.

Children

  • supervised learning — learn from labeled examples:
  • classification
  • regression
  • ranking
  • unsupervised learning — find structure without labels:
  • clustering
  • dimensionality reduction
  • anomaly detection
  • self-supervised learning — labels derived from the data itself (the basis of LLM pretraining)
  • semi-supervised learning — a little labeled, a lot unlabeled
  • reinforcement learning — learn a policy from reward:
  • agent · environment · state · action · reward · policy
  • probabilistic / Bayesian ML — reasoning under uncertainty