Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interaction as Explanation: A User Interaction-based Method for Explaining Image Classification Models

Hyeonggeun Yun

TL;DR

The paper tackles the interpretability problem in image classification by shifting from traditional feature-visualization to user-driven interaction. It introduces an interaction-based xAI framework implemented as a web-based prototype where users paint to mask image regions and observe resultant changes in class predictions and confidence, enabling end-users to compare their mental model with the model's reasoning. The approach is demonstrated on five diverse images using a ResNet-50 backbone, revealing how different image regions contribute to decisions and highlighting potential pathways for more intuitive, user-centric explainability. This work emphasizes end-user engagement and suggests future integration with existing xAI methods to broaden applicability across CV tasks while prompting user studies to assess understanding gains.

Abstract

In computer vision, explainable AI (xAI) methods seek to mitigate the 'black-box' problem by making the decision-making process of deep learning models more interpretable and transparent. Traditional xAI methods concentrate on visualizing input features that influence model predictions, providing insights primarily suited for experts. In this work, we present an interaction-based xAI method that enhances user comprehension of image classification models through their interaction. Thus, we developed a web-based prototype allowing users to modify images via painting and erasing, thereby observing changes in classification results. Our approach enables users to discern critical features influencing the model's decision-making process, aligning their mental models with the model's logic. Experiments conducted with five images demonstrate the potential of the method to reveal feature importance through user interaction. Our work contributes a novel perspective to xAI by centering on end-user engagement and understanding, paving the way for more intuitive and accessible explainability in AI systems.

Interaction as Explanation: A User Interaction-based Method for Explaining Image Classification Models

TL;DR

The paper tackles the interpretability problem in image classification by shifting from traditional feature-visualization to user-driven interaction. It introduces an interaction-based xAI framework implemented as a web-based prototype where users paint to mask image regions and observe resultant changes in class predictions and confidence, enabling end-users to compare their mental model with the model's reasoning. The approach is demonstrated on five diverse images using a ResNet-50 backbone, revealing how different image regions contribute to decisions and highlighting potential pathways for more intuitive, user-centric explainability. This work emphasizes end-user engagement and suggests future integration with existing xAI methods to broaden applicability across CV tasks while prompting user studies to assess understanding gains.

Abstract

In computer vision, explainable AI (xAI) methods seek to mitigate the 'black-box' problem by making the decision-making process of deep learning models more interpretable and transparent. Traditional xAI methods concentrate on visualizing input features that influence model predictions, providing insights primarily suited for experts. In this work, we present an interaction-based xAI method that enhances user comprehension of image classification models through their interaction. Thus, we developed a web-based prototype allowing users to modify images via painting and erasing, thereby observing changes in classification results. Our approach enables users to discern critical features influencing the model's decision-making process, aligning their mental models with the model's logic. Experiments conducted with five images demonstrate the potential of the method to reveal feature importance through user interaction. Our work contributes a novel perspective to xAI by centering on end-user engagement and understanding, paving the way for more intuitive and accessible explainability in AI systems.
Paper Structure (7 sections, 2 figures, 1 table)

This paper contains 7 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Interaction-based xAI
  3. Prototype for Interaction-based xAI Method
  4. Scenario of Interaction-based xAI Method
  5. Experiments
  6. Experiments Results
  7. Conclusions

Figures (2)

  • Figure 1: An example scenario of the interaction-based xAI method. 1) Our prototype displays the classification result of the initial image. 2) Users generate a masked image through the interaction. 3) The masked image is sent to the classification model to check the classification results. 4) The model sends the changed classification result back to the prototype. 5) The users could review the changed classification results. 6) The users could repeat the interaction process to deepen their understanding of the model through these interactions.
  • Figure 2: Experiment results of five images. The first column shows initial images. The second column shows images after the first interaction. The third column shows images after the second interaction. The classification results are shown below each image.