unsupervised

Definition

Learning without human-provided labels, relying on structure in the data itself — clustering and density estimation are the classical cases. The appeal is cost, since unlabeled data is abundant. The difficulty is evaluation: with no ground truth to score against, quality is usually judged by how well the learned representation transfers. Most modern "unsupervised" pre-training is really self-supervised, constructing its labels from the input.

Explainers using this term

• Paper Explained: Co-RL — Reasoning Without Labels, Emerging From a Diverse CohortCo-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL