A bi-directional adversarial explainability for decision support
Files
Accepted manuscript
Date
2021-02-20
Authors
Goldberg, Saveli
Pinsky, Eugene
Galitsky, Boris
Version
Accepted manuscript
OA Version
Citation
Saveli Goldberg, Eugene Pinsky, Boris Galitsky. 2021. "A Bi-directional Adversarial Explainability for Decision Support." Human-Intelligent Systems Integration, https://doi.org/10.1007/s42454-021-00031-5
Abstract
In this paper we present an approach to creating Bi-directional Decision Support System (DSS) as an intermediary between an expert (U) and a machine learning (ML) system for choosing an optimal solution. As a first step, such DSS analyzes the stability of expert decision and looks for critical values in data that support such a decision. If the expert’s decision and that of a machine learning system continue to be different, the DSS makes an attempt to explain such a discrepancy. We discuss a detailed description of this approach with examples. Three studies are included to illustrate some features of our approach.