PulseAugur
EN
LIVE 12:16:59

New methods improve Shapley value estimation for ML interpretability

Two new research papers propose novel methods for approximating Shapley values, a crucial metric for interpreting machine learning models. The first paper, "ShaplEIG," introduces a Bayesian experimental design approach that uses Gaussian processes to adaptively select coalitions for evaluation, improving sample efficiency in low-budget scenarios. The second paper, "An Odd Estimator for Shapley Values," reveals that Shapley values depend only on the odd component of set functions and proposes a new estimator, OddSHAP, which achieves state-of-the-art accuracy at larger sampling budgets by focusing on this odd subspace. AI

IMPACT These novel methods for Shapley value approximation could enhance the interpretability of complex machine learning models, particularly in resource-constrained settings.

RANK_REASON Two academic papers published on arXiv proposing new methods for approximating Shapley values.

Read on arXiv stat.ML →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

COVERAGE [2]

  1. arXiv stat.ML TIER_1 English(EN) · David Rundel, Fabian Fumagalli, Maximilian Muschalik, Bernd Bischl, Matthias Feurer ·

    ShaplEIG: Bayesian Experimental Design for Shapley Value Estimation

    arXiv:2606.02247v1 Announce Type: new Abstract: Shapley values are a principled attribution measure widely used in interpretable machine learning, but their exact computation scales exponentially with the number of players, motivating a wide range of approximation methods based o…

  2. arXiv stat.ML TIER_1 English(EN) · Fabian Fumagalli, Landon Butler, Justin Singh Kang, Kannan Ramchandran, R. Teal Witter ·

    An Odd Estimator for Shapley Values

    arXiv:2602.01399v2 Announce Type: replace-cross Abstract: The Shapley value is a ubiquitous framework for attribution in machine learning, encompassing feature importance, data valuation, and causal inference. However, its exact computation is generally intractable, necessitating…