AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Understanding Programmatic Weak Supervision via Source-aware Influence Function

Jieyu Zhang, Haonan Wang, Cheng-Yu Hsieh, Alexander Ratner

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use source-aware influence scores to identify LF mislabels instead of KNN discrepancy scoresPWS with Majority Voting · 13 datasets, Avg.AP: 0.640 -> 0.842Improved
Use source-aware reweighting influence instead of ordinary IF when selecting training components to removePWS with Majority Voting · 13 datasets, Avg.test loss: 0.660 -> 0.506Improved
Use source-aware reweighting influence instead of ordinary IF when selecting training components to removePWS with Dawid-Skene · 13 datasets, Avg.test loss: 0.648 -> 0.565Improved
Use source-aware reweighting RelatIF instead of ordinary RelatIF when selecting training components to removePWS with Snorkel · 13 datasets, Avg.test loss: 0.666 -> 0.499Improved
Use source-aware RelatIF instead of source-aware IF for LF mislabel identification (ablation)PWS with Majority Voting · 13 datasets, Avg.AP: 0.842 -> 0.881Improved
Use weight-moving rather than reweighting for source-aware IF (ablation)PWS · 13 datasets, Avg.test loss: MV: 0.506 vs 0.544; DS: 0.565 vs 0.588; Snorkel: 0.508 vs 0.559Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy