Research map / Label-efficient and robust learning / Recognition and long-tail
RealPatch: A Statistical Matching Framework for Model Patching with Real Samples
Sara Romiti, Christopher Inskip, Viktoriia Sharmanska, Novi Quadrianto
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 |
|---|---|---|---|
| add matched real-sample consistency regularisation (RealPatch vs SGDRO) | SGDRO · CelebA | robust accuracy (%): 82.96 -> 84.82 | Improved |
| use RealPatch instead of SGDRO | SGDRO · CelebA | robust gap (%): 7.13 -> 5.19 | Improved |
| use RealPatch instead of CAMEL | CAMEL · CelebA | robust gap (%): 5.09 -> 5.19 | No clear effect |
| use real matched pairs instead of CycleGAN counterfactuals (RealPatch vs CAMEL) | CAMEL · Waterbirds | robust accuracy (%): 76.82 -> 84.44 | Improved |
| use RealPatch instead of GDRO | GDRO · Waterbirds | aggregate accuracy (%): 88.26 -> 86.89 | Hurt |
| use RealPatch instead of GDRO | GDRO · Waterbirds | robust gap (%): 14.80 -> 4.43 | Improved |
| use RealPatch instead of SGDRO | SGDRO · Waterbirds | robust accuracy (%): 83.11 -> 84.44 | Improved |
| use RealPatch instead of ERM | ERM · iWildCam-small | robust gap (%): 19.65 -> 11.36 | Improved |
| use RealPatch instead of SGDRO | SGDRO · iWildCam-small | robust accuracy (%): 75.50 -> 76.70 | Improved |
| add temperature scaling (t<1) to propensity scores (ablation) | RealPatch Stage 1 matching · iWildCam-small covariate balance (SMD<=0.1 count) | covariates with SMD <= 0.1: 753 -> 1125 | Improved |
| add fixed caliper to matching (ablation) | RealPatch Stage 1 matching · iWildCam-small covariate balance (SMD<=0.1 count) | covariates with SMD <= 0.1: 1037 -> 1125 | Improved |
| match with RealPatch instead of original imSitu data | ResNet activity classifier · imSitu (211 classes) | model leakage: 76.79 -> 68.76 | Improved |
Closest papers by meaning
- Real-Fake: Effective Training Data Synthesis Through Distribution Matching (ICLR 2024 · similarity 0.84)
- Synthetic Tabular Data Generation for Imbalanced Classification: The Surprising Effectiveness of an Overlap Class (AAAI 2025 · similarity 0.83)
- Understanding Layer Patching in Model Size Interpolation (NeurIPS 2026 · similarity 0.83)
- Model Merging by Uncertainty-Based Gradient Matching (ICLR 2024 · similarity 0.82)
- Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging (NeurIPS 2025 · similarity 0.82)
- PLeaS - Merging Models with Permutations and Least Squares (CVPR 2025 · 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.
