Fair AI Plugin

Fair AI Plugin

In modern machine learning, interactive learning with multiple fairness metrics can significantly benefit both researchers and industry professionals in developing more equitable classifiers and models. By providing real-time insights and visualizations, such tools enable users to monitor fairness-related disparities during training, facilitating more informed decision-making. This project introduces a TensorBoard plugin designed to visualize fairness in ML models, starting with YOLOX and extending to broader applications. By integrating fairness metrics into the training workflow, this tool empowers users to identify and mitigate biases, ultimately fostering the development of fairer AI systems.

Key Results

• A TensorBoard-style plugin to visualize fairness indicators during model training.
• Integration tested with YOLOX for object detection on BDD and Cityscapes datasets.
• Real-time monitoring of fairness metrics across protected groups.
• Interactive dashboard for researchers to identify when and where unfairness arises.

Demo

Team

• Ray Chen — Ph.D Student (homepage)
• Prof. Christian Grant — Research Advisor (UF Profile)

Code and Data

• YOLOX Fairness Plugin (Private, contact project admin)
• Training Data: BDD100K, Cityscapes

Related Projects

• Fair AI Visulization

Sponsors

• NSF
• University of Florida Data Studio