Two good new reviews of XAI techniques have come out in the last week
Edin Hamzic from Symphony software has a new post “Overview of XAI methodologies” which summarizes research he did to pick the toolset for an XAI analysis of a tumor image recognition model that predicted the likelihood of malignancy. Certainly a high stakes use case which warrants an explanation. His paper focuses mostly on post-hoc interpretability alternatives and provides links to all the underlying research.
Christoph Molnar‘s e-book “Interpretable Machine Learning” is a very clearly written introduction to the field. Valuable both to practitioners and to non-Data Scientists with a good understanding of machine learning concepts. It includes useful context and descriptions of individual techniques.
The chapters on context cover: why interetability matters, what are the different types of explanations and what makes one explanation better than another. The chapters on techniques cover both interpretable models and model agnostic methods. He provides downloadable data sets to illustrate his points.
Chritoph’s book is being distributed with payment on the honor system, if you value the work he is providing please do contribute. Don’t be a cheapskate 😉