Research map / Image generation / Text-to-image diffusion
SurrogateSHAP: Training-Free Contributor Attribution for Text-to-Image (T2I) Models
MingYu Lu, Soham Gadgil, Chris Lin, Chanwoo Kim, Su-In Lee
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 proxy inference game instead of subset fine-tuning | SurrogateSHAP · CIFAR-20 proxy fidelity | time per subset query (min): 234.36 -> 10.72 | Improved |
| use proxy game instead of sFT | SurrogateSHAP · Fashion-Product proxy fidelity | RMS LPIPS: 0.337 -> 0.301 | Improved |
| use SurrogateSHAP instead of sFT Shapley | SurrogateSHAP · CIFAR-20 | LDS FID (%): -14.1 -> 51.21 | Improved |
| use SurrogateSHAP instead of sFT Shapley | SurrogateSHAP · ArtBench Post-Impressionism | LDS aesthetic (%): 18.85 -> 39.0 | Improved |
| use SurrogateSHAP instead of D-TRAK | SurrogateSHAP · ArtBench Post-Impressionism | LDS aesthetic (%): -8.75 -> 39.0 | Improved |
| use SurrogateSHAP tree Shapley instead of KernelSHAP at M=512 | SurrogateSHAP · synthetic interaction game | l2 error: 0.085 -> 0.010 | Improved |
| use SurrogateSHAP at low compute budget instead of training baseline | SurrogateSHAP · CIFAR-20 FID | LDS (%): 41.2% vs below 0% | Improved |
| remove bottom 20% groups chosen by SurrogateSHAP | SurrogateSHAP · ArtBench aesthetic | relative change (%): 3.62% | Improved |
| remove bottom 15% groups chosen by SurrogateSHAP | SurrogateSHAP · Fashion-Product diversity | relative change (%): 12.75% | Improved |
| remove top 10% groups chosen by SurrogateSHAP vs LOO | SurrogateSHAP · CIFAR-20 FID | FID increase (%): 31.70% vs 17.48% | Improved |
| remove top 20% groups chosen by SurrogateSHAP | SurrogateSHAP · ArtBench aesthetic | aesthetic drop (%): 7.71% | Improved |
Closest papers by meaning
- An Efficient Framework for Crediting Data Contributors of Diffusion Models (ICLR 2025 · similarity 0.89)
- Data Attribution for Text-to-Image Models by Unlearning Synthesized Images (NeurIPS 2024 · similarity 0.88)
- Diffusion Attribution Score: Evaluating Training Data Influence in Diffusion Models (ICLR 2025 · similarity 0.87)
- Fast Data Attribution for Text-to-Image Models (NeurIPS 2025 · similarity 0.86)
- FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models (CVPR 2024 · similarity 0.86)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.85)
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.
