2019

Generalized Integrated Gradients: A practical method for explaining diverse ensembles

Merrill, John, Ward, Geoff, Kamkar, Sean et al.

Understand

We introduce Generalized Integrated Gradients (GIG), a formal extension of the Integrated Gradients (IG) (Sundararajan et al., 2017) method for attributing credit to the input variables of a predictive model.

GIG improves IG by explaining a broader variety of functions that arise from practical applications of ML in domains like financial services.
GIG is constructed to overcome limitations of Shapley (1953) and Aumann-Shapley (1974), and has desirable properties when compared to other approaches.
We prove GIG is the only correct method, under a small set of reasonable axioms, for providing explanations for mixed-type models or games.

Built on

The Principles of Quantum Mechanics
Paul Dirac · 1930
Earlier work this paper cites.
A value for n-person games
Lloyd S. Shapley · 1953
Earlier work this paper cites.
Values of Non-Atomic Games
Rogert J. Aumann and Lloyd S. Shapley · 1974
Earlier work this paper cites.
Maximum likelihood estimation for mixed continuous and categorical data with missing values
Roderick J. A. Little and Mark D. Schluchter · 1985
Earlier work this paper cites.
Statistical Analysis with Missing Data
Roderick J A Little and Donald B Rubin · 1986
Earlier work this paper cites.
Learning representations by back propagating errors
David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams · 1986
Earlier work this paper cites.
Stacked generalization
David H. Wolpert · 1992
Earlier work this paper cites.

Similar

A parallel mixture of SVMs for very large scale problems
R. Collobert, S. Bengio, and Y. Bengio · 2002
Cited alongside, same era.
The Oxford Dictionary of Statistical Terms
Y. Dodge and International Statistical Institute · 2006
Cited alongside, same era.
Nonlinear boosting projections for ensemble construction
Nicolás García-Pedrajas, César García-Osorio, and Colin Fyfe · 2007
Cited alongside, same era.
Scikit-learn: Machine learning in Python
F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay · 2011
Cited alongside, same era.
TensorFlow: Large-scale machine learning on heterogeneous systems, 2015
Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mané, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viégas, Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng · 2015
Cited alongside, same era.
Keras, 2015
François Chollet et al · 2015
Cited alongside, same era.
XGBoost: A scalable tree boosting system
Original
Tianqi Chen and Carlos Guestrin · 2016
Cited alongside, same era.

Then

A unified approach to interpreting model predictions
Scott M Lundberg and Su-In Lee · 2017
Later among the works it cites.
Axiomatic attribution for deep networks
Original
Mukund Sundararajan, Ankur Taly, and Qiqi Yan · 2017
Later among the works it cites.
Consistent individualized feature attribution for tree ensembles
Original
Scott M Lundberg, Gabriel G Erion, and Su-In Lee · 2018
Later among the works it cites.
Systems and methods for decomposition of non-differentiable and differentiable models, 6 2018
Douglas C. Merrill, Michael Ruberry, Sean J. Kamkar, Jay Budzik, and John W. L. Merrill · 2018
Later among the works it cites.
SHAP (SHapley Additive exPlanations), 2019
Scott M Lundberg · 2019
Closest in time.
Systems and methods for decomposition of non-differentiable and differentiable models, 2 2019
John W. L. Merrill, John J. Beahan, Geoffrey M. Ward, Sean J. Kamkar, Jay Budzik, Jose E. Valentin, Mark F. Eberstein, and Douglas C. Merrill · 2019
Closest in time.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…

Reading the bibliography…