Research map / Learning theory and algorithms / Prediction and decision losses
Shapley value attribution
Estimate Shapley values efficiently for feature attribution and data valuation.
63 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Prediction and decision losses, keyword cluster “explanations · shapley · attribution”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
18 papers in 2023–24 and 27 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.4% to 4.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Using the Shapley Value to Analyze Algorithm Portfolios (AAAI 2016 · 26 citations)
- 2019 · Data Shapley: Equitable Valuation of Data for Machine Learning (ICML 2019 · 147 citations)
- 2020 · Problems with Shapley-value-based explanations as feature importance measures (ICML 2020 · 110 citations)
- 2021 · On the Tractability of SHAP Explanations (AAAI 2021 · 471 citations)
- 2022 · RKHS-SHAP: Shapley Values for Kernel Methods (NeurIPS 2022 · 5 citations)
- 2023 · Faith-Shap: The Faithful Shapley Interaction Index (ICML 2023 · 19 citations)
- 2024 · Shaping Up SHAP: Enhancing Stability through Layer-Wise Neighbor Selection (AAAI 2024 · 18 citations)
- 2025 · Provably Accurate Shapley Value Estimation via Leverage Score Sampling (ICLR 2025 · 3 citations)
- 2026 · Verified SHAP: Provable Bounds for Exact Shapley Values of Neural Networks (ICML 2026 · 0 citations)
Earliest papers here
- Using the Shapley Value to Analyze Algorithm Portfolios (AAAI 2016 · 26 citations)
- Data Shapley: Equitable Valuation of Data for Machine Learning (ICML 2019 · 147 citations)
- Explaining Deep Neural Networks with a Polynomial Time Algorithm for Shapley Value Approximation (ICML 2019 · 47 citations)
What has been tried
47 papers here report 387 measured changes (a component added, removed or swapped, compared with a baseline): helped 69%, no clear effect 9%, hurt 14%, mixed 8%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| increase explanation order k from 1 (Shapley value) to higher orders | synthetic and ML applications · Shapley-weighted R^2 | Helped | shapiq: Shapley Interactions for Machine Learning (NeurIPS 2024 · 13 citations) | |
| Add paired sampling | Sample-size experiments across settings · relative squared ℓ2 error | roughly four or more | Helped | Provably Accurate Shapley Value Estimation via Leverage Score Sampling (ICLR 2025 · 3 citations) |
| Use Precompute instead of Base implementation | Implementation experiments · runtime | Helped | SHAP values via sparse Fourier representation (NeurIPS 2025 · 2 citations) | |
| Replace the linear approximation with a tree-based approximation | Eight datasets, m=40n · relative squared ℓ2 error | 5.35e-04 -> 1.35e-04 | Helped | Regression-adjusted Monte Carlo Estimators for Shapley Values and Probabilistic Values (NeurIPS 2025 · 1 citation) |
| vary the RNA base encoding scheme | RNA splicing efficiency reconstruction · R² | 0.80, 0.77, 0.78, 0.77 | No clear effect | SHAP zero Explains Biological Sequence Models with Near-zero Marginal Cost for Future Queries (NeurIPS 2025 · 1 citation) |
| Use a first-order Taylor approximation | GPT2, Pile, 3500th training iteration · RMSE against Monte Carlo estimate | 0.0003 | Helped | Data Shapley in One Training Run (ICLR 2025 · 1 citation) |
| Use GEM-FIX-LR instead of the regularized GEM-FIX formulation | Cooperative-game simulation · Shapley-value estimation availability | Hurt | Unlocking the Game: Estimating Games in Möbius Representation for Explanation and High-Order Interaction Detection (AAAI 2025 · 0 citations) | |
| Increase the number of background samples from 50 to 100 | OpenML-CTR23 and OpenML-CC18; functional component importance · datasets with relative differences exceeding ±30% | 18 -> 9 | Helped | Fast Estimation of Partial Dependence Functions using Trees (ICML 2025 · 0 citations) |
| Train the KAN with Shapley loss | Tabular prediction · AUC | Helped | Prediction via Shapley Value Regression (ICML 2025 · 0 citations) | |
| sample without replacement instead of with replacement | evaluated datasets · normalized MSE | No clear effect | A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values (NeurIPS 2025 · 0 citations) | |
| Increase patch set from 1×1 to 2×2 | CUB fine-grained · accuracy | 33.2% -> 45.8% | Helped | InstaSHAP: Interpretable Additive Models Explain Shapley Values Instantly (ICLR 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 22 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Adult · 3 of 22 papers
- breast cancer · 2 of 22 papers
- diabetes · 2 of 22 papers
- CIFAR-10 · 2 of 22 papers
- Bike Sharing · 2 of 22 papers
- Independent · 2 of 22 papers
- NHANES · 2 of 22 papers
- Communities · 2 of 22 papers
Baselines
- Kernel SHAP · 10 of 22 papers
- FastSHAP · 4 of 22 papers
- LeverageSHAP · 3 of 22 papers
- LIME · 2 of 22 papers
- Bivariate SHAP · 2 of 22 papers
- SHAP · 2 of 22 papers
- Sampling SHAP · 2 of 22 papers
- pltreeshap · 2 of 22 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 45 papers; from OpenAlex and the papers' affiliations).
- Ludwig-Maximilians-Universität München · 4 papers
- Universität Paderborn · 3 papers
- Universität Bielefeld · 3 papers
- LMU Munich · 2 papers
- Inria · 2 papers
- Vrije Universiteit Amsterdam · 2 papers
- University of Chicago · 2 papers
- Stanford University · 2 papers
- East China Normal University · 2 papers
- KTH Royal Institute of Technology · 2 papers
Use it
Download BibTeX of 36 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Provable post-hoc explanations (92 papers)
- Counterfactual explanations and recourse (46 papers)
Elsewhere in the topic
- Adaptive testing and experiments (201 papers)
- Active learning and subset selection (199 papers)
- Distributionally robust optimization (184 papers)
- Data attribution and recourse (160 papers)
- Theory of neural network sensitivity (143 papers)
- Ranking models from comparisons (127 papers)
- Determinantal point process inference (122 papers)
- Conformal prediction sets (116 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).