A1589
Title: Priority-aware Shapley value
Authors: Yuan Zhang - The Ohio State University (United States) [presenting]
Abstract: Shapley values are widely used for model-agnostic data valuation and feature attribution, yet they implicitly assume contributors are interchangeable. This can be problematic when contributors are dependent (e.g., reused or augmented data or causal feature orderings) or when contributions should be adjusted by factors such as trust or risk. Priority-Aware Shapley Value (PASV) incorporates both hard precedence constraints and soft, contributor-specific priority weights. PASV is applicable to general precedence structures, recovers precedence-only and weight-only Shapley variants as special cases, and is uniquely characterized by natural axioms. An efficient adjacent-swap Metropolis-Hastings sampler enables scalable Monte Carlo estimation, and limiting regimes induced by extreme priority weights are analyzed. Experiments on data valuation (MNIST and CIFAR10) and feature attribution (Census Income) demonstrate more structure-faithful allocations and a practical sensitivity analysis via priority sweeping.
