Reproduce “Snorkel: Rapid Training Data Creation with Weak Supervision” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alexander Ratner, Stephen H. Bach, Henry Ehrenberg, Jason Fries, Sen Wu, Christopher Ré
Snorkel, from Stanford University's Hazy Research lab, provides an end-to-end system for rapidly creating training data without manual labeling. It leverages programmatic labeling functions and a generative model to denoise weak supervision signals, achieving performance comparable to hand-supervised models across diverse applications.
Key insights
- "Data programming" allows domain experts to encode knowledge programmatically as Labeling Functions (LFs), shifting the bottleneck from manual data labeling to logic specification.
- A probabilistic generative model can effectively denoise and combine conflicting weak supervision signals from LFs to create high-quality, probabilistic training labels without requiring ground truth.
- A discriminative model trained on these probabilistic labels can learn to generalize beyond the specific heuristics encoded in the LFs, leading to improved coverage and robust performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.