loading page

Generating and Evaluating Explanations of Attended and Error-Inducing Input Regions for VQA Models
  • +4
  • Arijit Ray,
  • Michael Cogswell,
  • Xiao Lin,
  • Kamran Alipour,
  • Ajay Divakaran,
  • Yi Yao,
  • Giedrius Burachas
Arijit Ray
SRI International

Corresponding Author:arijit.ray93@gmail.com

Author Profile
Michael Cogswell
SRI International
Author Profile
Xiao Lin
SRI International
Author Profile
Kamran Alipour
University of California San Diego
Author Profile
Ajay Divakaran
SRI International
Author Profile
Yi Yao
SRI International
Author Profile
Giedrius Burachas
SRI International
Author Profile

Abstract

Attention maps, a popular heatmap-based explanation method for Visual Question Answering (VQA), are supposed to help users understand the model by highlighting portions of the image/question used by the model to infer answers. However, we see that users are often misled by current attention map visualizations that point to relevant regions despite the model producing an incorrect answer. Hence, we propose Error Maps that clarify the error by highlighting image regions where the model is prone to err. Error maps can indicate when a correctly attended region may be processed incorrectly leading to an incorrect answer, and hence, improve users’ understanding of those cases. To evaluate our new explanations, we further introduce a metric that simulates users’ interpretation of explanations to evaluate their potential helpfulness to understand model correctness. We finally conduct user studies to see that our new explanations help users understand model correctness better than baselines by an expected 30% and that our proxy helpfulness metrics correlate strongly (rho>0.97) with how well users can predict model correctness.
20 Jun 2021Submitted to Applied AI Letters
21 Jun 2021Submission Checks Completed
21 Jun 2021Assigned to Editor
25 Jun 2021Reviewer(s) Assigned
07 Jul 2021Review(s) Completed, Editorial Evaluation Pending
26 Jul 2021Editorial Decision: Revise Minor
24 Aug 20211st Revision Received
25 Aug 2021Submission Checks Completed
25 Aug 2021Assigned to Editor
13 Sep 2021Reviewer(s) Assigned
11 Oct 2021Review(s) Completed, Editorial Evaluation Pending
25 Oct 2021Editorial Decision: Accept
24 Nov 2021Published in Applied AI Letters. 10.1002/ail2.51