• Publications
  • Influence
A Unified Approach to Interpreting Model Predictions
TLDR
A unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations), which unifies six existing methods and presents new methods that show improved computational performance and/or better consistency with human intuition than previous approaches. Expand
From local explanations to global understanding with explainable AI for trees
TLDR
An explanation method for trees is presented that enables the computation of optimal local explanations for individual predictions, and the authors demonstrate their method on three medical datasets. Expand
Consistent Individualized Feature Attribution for Tree Ensembles
TLDR
This work develops fast exact tree solutions for SHAP (SHapley Additive exPlanation) values, which are the unique consistent and locally accurate attribution values, and proposes a rich visualization of individualized feature attributions that improves over classic attribution summaries and partial dependence plots, and a unique "supervised" clustering. Expand
Explainable machine-learning predictions for the prevention of hypoxaemia during surgery
TLDR
The results suggest that if anaesthesiologists currently anticipate 15% of hypoxaemia events, with the assistance of this system they could anticipate 30%, a large portion of which may benefit from early intervention because they are associated with modifiable factors. Expand
Explainable AI for Trees: From Local Explanations to Global Understanding
TLDR
Improvements to the interpretability of tree-based models through the first polynomial time algorithm to compute optimal explanations based on game theory, and a new type of explanation that directly measures local feature interaction effects. Expand
Consistent feature attribution for tree ensembles
TLDR
This work develops fast exact solutions for SHAP (SHapley Additive exPlanation) values, which were recently shown to be the unique additive feature attribution method based on conditional expectations that is both consistent and locally accurate, and demonstrates the inconsistencies of current methods. Expand
Understanding Global Feature Contributions With Additive Importance Measures
TLDR
This work proposes SAGE, a model-agnostic method that quantifies predictive power while accounting for feature interactions and shows that SAGE can be calculated efficiently and that it assigns more accurate importance values than other methods. Expand
Learning Explainable Models Using Attribution Priors
TLDR
A differentiable axiomatic feature attribution method called expected gradients is developed and shown how to directly regularize these attributions during training and produce models with more intuitive behavior and better generalization performance by encoding constraints that would otherwise be very difficult to encode using standard model priors. Expand
Understanding Global Feature Contributions Through Additive Importance Measures
TLDR
This work proposes a new feature importance method, Shapley Additive Global importancE (SAGE), a model-agnostic measure of feature importance based on the predictive power associated with each feature, which relates to prior work through the novel framework of additive importance measures. Expand
...
1
2
3
4
...