Research map / Learning theory and algorithms / Clustering
Error Discovery By Clustering Influence Embeddings
Fulton Wang, Julius Adebayo, Sarah Tan, Diego Garcia-Olano, Narine Kokhlikyan
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use InfEmbed instead of Domino | slice discovery · dcbench natural images, rare | precision@10: 0.4 -> 0.65 | Improved |
| use InfEmbed instead of Domino | slice discovery · dcbench natural images, correlation | precision@10: 0.45 -> 0.55 | Improved |
| use InfEmbed instead of Domino | slice discovery · dcbench medical images, rare | precision@10: 0.39 -> 0.57 | Improved |
| use InfEmbed instead of Domino | slice discovery · dcbench medical time series, noisy label | precision@10: 0.9 -> 0.81 | Hurt |
| use InfEmbed instead of Domino | slice discovery · dcbench | coherence score: 1219 out of 1235 | Improved |
| use InfEmbed instead of Domino | slice discovery · dcbench | label homogeneity | Mixed |
| use InfEmbed instead of PlaneSpot | slice discovery · SpotCheck | discovery rate: 0.85 -> 0.91 | Improved |
| use InfEmbed instead of PlaneSpot | slice discovery · SpotCheck | false discovery rate: 0.07 -> 0.15 | Hurt |
| replace K-Means with DBSCAN (ablation) | InfEmbed · SpotCheck | discovery rate: 0.91 -> 0.92 | No clear effect |
| replace K-Means with spectral clustering (ablation) | InfEmbed · SpotCheck | discovery rate: 0.91 -> 0.89 | No clear effect |
| branching factor B=2 vs B=3 (ablation) | InfEmbed-Rule · dcbench | slice quality | No clear effect |
| use influence embeddings (gradient-based) instead of last-layer representations only | slice discovery · analysis | slice similarity | Improved |
Closest papers by meaning
- HiBug2: Efficient and Interpretable Error Slice Discovery for Comprehensive Model Debugging (ICLR 2025 · similarity 0.83)
- The Mirrored Influence Hypothesis: Efficient Data Influence Estimation by Harnessing Forward Passes (CVPR 2024 · similarity 0.82)
- Most Influential Subset Selection: Challenges, Promises, and Beyond (NeurIPS 2024 · similarity 0.82)
- Data-centric Prediction Explanation via Kernelized Stein Discrepancy (ICLR 2025 · similarity 0.82)
- Foundations of Testing for Finite-Sample Causal Discovery (ICML 2024 · similarity 0.82)
- Data-Efficient Learning via Clustering-Based Sensitivity Sampling: Foundation Models and Beyond (ICML 2024 · similarity 0.82)
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.
