An Integrated Approach to Explainable Machine Learning

Abstract

Explainable Machine Learning (ML) has recently gained a great attention due to its significant role in helping users better understand how ML models work so as to improve their trust in ML applications. This project aims to investigate the role and the integration of reasoning mechanisms using Argumentation theory and Dempster-Shafer theory of evidence (evidential reasoning) in developing a novel comprehensive framework for evolving explainable learning systems, articulated along the following research challenges: 1) how to generate the most faithful explanation of a given (black-box) learning model; 2) how to design inherently interpretable models without the cost of sacrificing accuracy for interpretability; 3) how black-box models with their explanations and inherently interpretable models collaborate and push each other to evolve their capabilities for both accuracy and interpretability. As a novel test-bed, we shall deploy the methods and tools developed in an argumentation-basedexplainable ML platform. The effectiveness and applicability of the proposed framework will be demonstrated by applications in the areas of biomedical research and recommender systems.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 24, 2023
Source ID
N629092312058

Entities

People

  • Bac Le

Organizations

  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Distributed Systems and Data Platform Development
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks
  • Biotechnology