Designing for Visualising Multiple Counterfactuals for XAI

Problem Statement

Machine learning models are prevalent, but their "black box" nature hinders user understanding. Counterfactuals, alternative scenarios that change model predictions, offer a path forward. This thesis focuses on building a visualization tool to efficiently present multiple counterfactuals. While existing research explores generating counterfactuals, this work delves into visualization. We leverage a user-friendly web-based tool employing tSNE to showcase multiple counterfactuals. The tool empowers users to explore "what-if" scenarios, fostering interpretability and understanding for non-technical audiences. Through user studies, we assess the effectiveness of our visualization tool. This thesis aims to bridge the gap between machine learning models and end-users, enhancing decision-making through clear and accessible explanations.

Tasks

• Perform a literature review
• Familiarize yourself with developing visualization tools
• Implement an interactive website for the users
• Conduct a user study
• Do a statistical evaluation
• Write a thesis and present your findings in the Disputationsseminar
• (Optional:) co-write a research paper