M31.7 CONNECT THE MECHANISM
Use unlabeled data and imperfect label sources carefully
A marketplace has 1,000 labeled listings and 50,000 unlabeled ones. Turn confident guesses and rough rules of thumb into training labels, without letting the model believe its own mistakes.
LESSON OVERVIEW14 min lesson
Lesson overview
A marketplace has 1,000 labeled listings and 50,000 unlabeled ones. Turn confident guesses and rough rules of thumb into training labels, without letting the model believe its own mistakes.
What you’ll explore
- Train on confident pseudo-labels with a threshold, explain consistency regularization as in FixMatch, combine noisy labeling-function votes by their accuracies, and check every label source against a hand-labeled gold set.
GO TO THE SOURCE
Original explanations, connected to the research.
FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (Sohn et al., 2020)Snorkel: Rapid Training Data Creation with Weak Supervision (Ratner et al., 2017)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.