Fair AI Project
By Ray Chen on
Jun 30, 2025
Fair AI Visulization
The Fair AI project introduces an interactive visualization system for evaluating fairness in AI models using Signed Deviation Error (S.D.E.). The tool enables researchers to interpret bias in predictive systems across sensitive (e.g., weather) and domain (e.g., time of day) attributes. Using median-based signed error shifts, the visualization highlights disparities in subgroup performance, offering intuitive insight into model behavior.
Key Results
- Introduced the new metric (median deviation of signed error) as a fairness indicator.
- Developed an interactive dashboard to visualize group-level error dynamics.
- Demonstrated subgroup bias patterns using BDD-based autonomous driving data.
- Embedded subgroup clustering using t-SNE and TRIMAP to reveal structural bias.
Team
- Ray Chen — Ph.D Student
- Christopher William Driggers-Ellis - Ph.D Student
- Prof. Christian Grant – Advisor
Code and Data
- Fedora Visualization (Private Repo)
- Datasets: BDD100K, with weather/time-of-day annotations.
- Demo Visulization Repo
Publications
Come out in this Fall!
