Explainable Neural Networksby Michael Kubiske
Director, Center for Machine Learning NYC
This talk will run through an overview of current methodologies for neural network explain-ability and tie them back into the financial services space. It will also highlight recent advancements and some of our research in the field with a brief discussion of our working paper for use in heavily regulated model builds. Included in our areas of interest are the generation of baseline vectors in a model specific context as opposed to model agnostic concepts like LIME. The talk will close with a brief explanation of why we explore model specific methodologies for regulatory reasons and their potential defensibility to regulators.