DataLabeling

Advanced labeling techniques * Active learning sampling techniques

• Maargin sampling

  • Label samples that the model is the least confident with

• Cluster-based sampling

  • sample from cluster to cover the entire space

• Query by committee

  • Train an ensemble of models and sample points that are in disagreement

• Region-based sampling

  • Runs several active learning algorithms in different parts of the space

• Weak Supervision https://www.snorkel.org/