Advances in Explainable AI and Data Valuation

The field of explainable AI and data valuation is rapidly moving towards developing more robust and efficient methods for attributing feature importance and evaluating data contributions. Researchers are focusing on addressing the challenges of misreporting, hyperparameter sensitivity, and scalability in existing methods. New frameworks and algorithms are being proposed to provide more accurate and reliable explanations, such as multi-criteria rank-based aggregation and causally-motivated approaches. Additionally, there is a growing interest in applying Gaussian processes and Shapley values to data valuation and feature attribution. Notable papers in this area include the proposition of a unified framework for provably efficient algorithms to estimate Shapley values, and the introduction of Fast-DataShapley, a one-pass training method for training data valuation. These innovations are expected to significantly advance the field and enable more transparent and trustworthy AI systems.

Sources

Estimating Misreporting in the Presence of Genuine Modification: A Causal Perspective

Taming Hyperparameter Sensitivity in Data Attribution: Practical Selection Without Costly Retraining

Multi-criteria Rank-based Aggregation for Explainable AI

Feature Attribution from First Principles

On the Usage of Gaussian Process for Efficient Data Valuation

MVP-Shapley: Feature-based Modeling for Evaluating the Most Valuable Player in Basketball

A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values

Fast-DataShapley: Neural Modeling for Training Data Valuation

Built with on top of