Faithful Group Shapley Value
NeurIPS 2025, 2025
Faithful Group Shapley Value addresses group-level data valuation when data providers contribute batches of data. The paper identifies a strategic group-splitting vulnerability in existing approaches and proposes a faithful valuation method with efficient approximation algorithms.
Recommended citation: Lee, K., Liu, Z., Tang, W., and Zhang, Y. (2025). "Faithful Group Shapley Value." Advances in Neural Information Processing Systems (NeurIPS). ICML 2025 DataWorld Workshop Best Paper Award Honorable Mention.
Download Paper