Research map / Learning theory and algorithms / Submodular and game algorithms
Rethinking Shapley Value for Negative Interactions in Non-convex Games
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Revisits Shapley-value-based payoff allocation to explicitly account for interactions between players, including negative interactions, in cooperative games for feature attribution.
Problem
Shapley-value-based feature attribution for deep learning does not explicitly capture interactions between players (features), particularly negative interactions in non-convex games.
Approach
Rethinks the Shapley value criterion so that causal interactions between players are represented explicitly in payoff allocation, with specific treatment of negative interactions in non-convex games.
Experimental setup
Builds on: Shapley value
Closest papers by meaning
- Approximating the Shapley Value without Marginal Contributions (AAAI 2024 · similarity 0.88)
- Unlocking the Game: Estimating Games in Möbius Representation for Explanation and High-Order Interaction Detection (AAAI 2025 · similarity 0.88)
- shapiq: Shapley Interactions for Machine Learning (NeurIPS 2024 · similarity 0.87)
- Explaining Probabilistic Models with Distributional Values (ICML 2024 · similarity 0.87)
- KernelSHAP-IQ: Weighted Least Square Optimization for Shapley Interactions (ICML 2024 · similarity 0.87)
- Towards Attributions of Input Variables in a Coalition (ICML 2025 · 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.