Research map / Probabilistic and scientific modeling / Molecules and proteins
An Odd Estimator for Shapley Values
Fabian Fumagalli, Landon Butler, Justin Singh Kang, Kannan Ramchandran, R. Teal Witter
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 |
|---|---|---|---|
| restrict regression to odd Fourier interactions with paired sampling (ablation) | OddSHAP · Cancer, 10,000 samples | MSE ratio vs LeverageSHAP: up to 62x error reduction | Improved |
| include about 1,000 interactions (eta = 10) (ablation) | OddSHAP · all value functions, 10,000 samples | MSE ratio vs LeverageSHAP: at least 6x reduction | Improved |
| include too many interactions (smaller eta) (ablation) | OddSHAP · all value functions, 10,000 samples | MSE ratio | Hurt |
| use non-paired instead of paired sampling (ablation) | OddSHAP · all datasets, 10,000 samples | MSE ratio vs OddSHAP | Hurt |
| add even interactions under paired sampling (All instead of Odd) (ablation) | OddSHAP · all datasets, 10,000 samples | MSE ratio vs OddSHAP | Hurt |
| OddSHAP vs RegressionMSR on low-dimensional value functions | OddSHAP · DistilBERT, Estate, ViT16 | MSE | No clear effect |
| OddSHAP vs flexible-budget baselines with sufficient budget | OddSHAP · Cancer, IL60, CG60, NHANES, Crime | MSE: Crime 5.6x10^-1 -> 1.3x10^-1 | Improved |
| OddSHAP vs FFD-RD on tree-based value functions | OddSHAP · Estate, Cancer, IL60, CG60 | MSE: Cancer 4.2x10^-6 vs 6.4x10^-7 | Hurt |
| OddSHAP vs FFD-RD on deep-learning value functions | OddSHAP · DistilBERT, ViT16 | MSE: DistilBERT 1.6x10^-3 -> 4.6x10^-5 | Improved |
| add adaptive bias correction to FFD-RD | FFD-RD · most value functions | MSE: DistilBERT 1.6x10^-3 -> 1.8x10^-3 | Hurt |
Closest papers by meaning
- A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values (NeurIPS 2025 · similarity 0.91)
- DU-Shapley: A Shapley Value Proxy for Efficient Dataset Valuation (NeurIPS 2024 · similarity 0.88)
- Provably Accurate Shapley Value Estimation via Leverage Score Sampling (ICLR 2025 · similarity 0.88)
- Regression-adjusted Monte Carlo Estimators for Shapley Values and Probabilistic Values (NeurIPS 2025 · similarity 0.87)
- Amortized Linear-time Exact Shapley Value for Product-Kernel Methods (NeurIPS 2026 · similarity 0.87)
- shapiq: Shapley Interactions for Machine Learning (NeurIPS 2024 · similarity 0.87)
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.
