Model Explainability Research
References from Susmit Jha
- Light reading: https://www.theatlantic.com/technology/archive/2018/01/equivant-compas-algorithm/550646/
- Associated dataset: https://www.kaggle.com/danofer/compass
- A book by Moritz et al. https://fairmlbook.org/pdf/fairmlbook.pdf (not really to read through it but rather to look at it to get a feel of the area)
- What experts in law think about this: https://www.nyulawreview.org/wp-content/uploads/2019/06/NYULawReview-94-3-ODonnell.pdf
There are different attack vectors (not necessarily orthogonal) on this problem:
- Causality: http://bayes.cs.ucla.edu/BOOK-2K/book-toc.html (again just take a look to see if this interests you)
- Shapley values https://christophm.github.io/interpretable-ml-book/shapley.html (just this chapter of the book)
- Representation learning https://towardsdatascience.com/disentanglement-with-variational-autoencoder-a-review-653a891b69bd
- Mutual information https://medium.com/towards-artificial-intelligence/mine-mutual-information-neural-estimation-26f6853febda
Additional Resources to Investigage
- IBM
- https://github.com/Trusted-AI/AIX360
- AIX360 open source project has joined the Linux Foundation
- https://github.com/IBM/lale
- built on sklearn
- https://github.com/Trusted-AI/AIX360