Activities per year
Abstract
The Shapley Additive Global Importance (SAGE) value is a theoretically appealing interpretability method that fairly attributes global importance to a model?s surplus performance contributions over an exponential number of feature sets. This is computationally expensive, particularly because estimating the surplus contributions requires sampling from conditional distributions. Thus, SAGE approximation algorithms only take a fraction of the feature sets into account. We propose dSAGE, a method that accelerates SAGE approximation. dSAGE is motivated by the observation that conditional independencies (CIs) between a feature and the model target imply zero surplus contributions, such that their computation can be skipped. To identify CIs, we leverage causal structure learning (CSL) to infer a graph that encodes (conditional) independencies in the data as dseparations. This is computationally more efficient because the expense of the onetime graph inference and the dseparation queries is negligible compared to the expense of surplus contribution evaluations. Empirically we demonstrate that dSAGE enables the efficient and accurate estimation of SAGE values.
Original language  English 

Pages (fromto)  1165011670 
Number of pages  21 
Journal  Proceedings of Machine Learning Research (PMLR) 
Volume  206 
Publication status  Published  25 Apr 2023 
Austrian Fields of Science 2012
 102019 Machine learning
Fingerprint
Dive into the research topics of 'Efficient SAGE Estimation via Causal Structure Learning'. Together they form a unique fingerprint.Activities
 1 Poster presentation

Efficient SAGE Estimation via Causal Structure Learning
Christoph Luther (Speaker)
25 Apr 2023 → 27 Apr 2023Activity: Talks and presentations › Poster presentation › Science to Science