Unlabeled pool Labeled class 0 Labeled class 1 QBC accuracy Random accuracy
⚠ Couldn't renderCanvas 2D failed to initialize. Reload the page.

Query-by-Committee: Vote-Entropy Active Learning (2D)

Active learning tries to reach high accuracy from as few labeled examples as possible by choosing, at every step, which unlabeled point is worth asking a human to label. This simulator implements Query-by-Committee (QBC): a committee of logistic-regression classifiers is trained on bootstrap resamples of the labeled set, every unlabeled pool point gets a vote from each member, and the point with the highest vote entropy — the one the committee disagrees about most — is queried next. A random-query baseline runs alongside it from the same starting labels, and a live accuracy-vs-labels chart shows exactly how much the disagreement-driven strategy saves.