Priority-Aware Shapley Value

Priority-Aware Shapley Value (PASV) extends Shapley-based attribution to settings where contributors are not fully interchangeable. The method incorporates both hard precedence constraints and soft priority weights, with applications to data valuation and feature attribution.

Recommended citation: Lee, K., Liu, Z., Tang, W., and Zhang, Y. (2026). "Priority-Aware Shapley Value." International Conference on Machine Learning (ICML).
Download Paper