Explainable AI to support Veterans Transition Assistance Programs (XAI-VTAP)

Award Information
Agency: Department of Defense
Branch: Defense Advanced Research Projects Agency
Contract: 140D6318C0053
Agency Tracking Number: D2-2053
Amount: $1,499,184.00
Phase: Phase II
Program: SBIR
Solicitation Topic Code: SB163-007
Solicitation Number: 2016.3
Timeline
Solicitation Year: 2016
Award Year: 2018
Award Start Date (Proposal Award Date): 2018-06-05
Award End Date (Contract End Date): 2019-06-05
Small Business Information
3600 Green Court, Ann Arbor, MI, 48105
DUNS: 009485124
HUBZone Owned: N
Woman Owned: N
Socially and Economically Disadvantaged: N
Principal Investigator
 Dr. Scott Lathrop
 (734) 845-2568
 scott.lathrop@soartech.com
Business Contact
 Laura Schwennesen
Phone: (734) 877-7683
Email: proposals@soartech.com
Research Institution
N/A
Abstract
A challenge for Artificial Intelligence recommenders is explaining their reasoning to users. Humans tend to explain concepts with symbolic narratives, which contrasts the way non-symbolic machine learning represents concepts. This research addresses this challenge by grounding the research in the problem of helping transitioning veterans find relevant careers. The developed prototype will provide job recommendations emerging from a Long Short-Term Memory trained on word embeddings derived from 1.73 million job descriptions and 70,000 resumes. Explanations of matches are grounded in selected text segments, or StoryGraph elements, from an individuals resume with a skill decomposition graph providing additional clarification. By grounding the explanation, veterans will be able to understand causal relationships between resumes and job descriptions. The proposed second phase will investigate conceptual explanation through hierarchical learning and the potential for attention-based neural networks to improve the matching and semantic explanations. We will also extend the StoryGraph to provide insights into career path trajectories and create an interactive interface for veterans to gain deeper insights and explore hypotheticals. We propose an option task to explore generalizability by extending it to the medical domain to help match doctors and patients.

* Information listed above is at the time of submission. *

Agency Micro-sites

SBA logo
Department of Agriculture logo
Department of Commerce logo
Department of Defense logo
Department of Education logo
Department of Energy logo
Department of Health and Human Services logo
Department of Homeland Security logo
Department of Transportation logo
Environmental Protection Agency logo
National Aeronautics and Space Administration logo
National Science Foundation logo
US Flag An Official Website of the United States Government