Researchers have developed OperatorSHAP, a novel method for estimating Shapley values in neural operators. This approach addresses the computational cost and input limitations of existing explainability techniques like FastSHAP, particularly for applications involving irregular data grids. OperatorSHAP provides a grid-agnostic attribution method and a training procedure that connects to Aumann-Shapley values, demonstrating consistency with discrete Shapley values across resolutions and grid sizes without retraining. AI
IMPACT Improves explainability for neural operators, crucial for safety-critical physical applications.
RANK_REASON The cluster contains a research paper detailing a new method for explainability in machine learning.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
