AAAI Publications, 2018 AAAI Spring Symposium Series

Font Size: 
Revealing Actionable Simplicity in Data
Nick Gisolfi, Artur Dubrawski

Last modified: 2018-03-15

Abstract


We present a methodology where we identify simple structure in data, if such structure exists, and present it to end-users, enabling them to interact with data or manipulate a machine learning model. We share our bounding box algorithm which distills complex information into a small set of range rules which yield naturally intuitive visualizations. We demonstrate a few cases where simple, actionable descriptions lead to quantitative improvements in an AI pipeline.

Keywords


explainable;AI;bounding;box;algorithm;user;interpretable;ML;structure;pattern;search;detection

Full Text: PDF