Priority-Aware Shapley Value
Researchers have introduced Priority-Aware Shapley Value (PASV), a novel method for data valuation and feature attribution that addresses the limitations of traditional Shapley values. PASV incorporates precedence constraints and contributor-specific priority weights, allowing for more nuanced allocation of contributions. The method is characterized by natural axioms and an efficient Metropolis-Hastings sampler for scalable estimation. Experiments on datasets like MNIST, CIFAR-10, and Census Income demonstrate PASV's ability to produce more structure-faithful allocations and enable practical sensitivity analysis. AI
IMPACT Introduces a more nuanced approach to AI data valuation and feature attribution, potentially improving model interpretability and fairness.