Yongkai Wu Research Focus

Research Focus: Dr. Yongkai Wu
Assistant Professor, Electrical and Computer Engineering
Clemson University

FairAgent: Making Fair Machine Learning Accessible to All

Dr. Yongkai Wu is an Assistant Professor in the Department of Electrical and Computer Engineering at Clemson University. His research focuses on fairness-aware machine learning, explainable artificial intelligence, and causal inference. He is a recipient of the GAIN (Grants for Applications in Industry and Networking) award from the ADAPT in SC Project.

Automated Tools for Fairness-Aware Machine Learning

Machine learning systems increasingly influence critical decisions in healthcare, education, and criminal justice. However, these systems can perpetuate or amplify societal biases, leading to discriminatory outcomes. Developing fair machine learning models requires deep expertise in fairness metrics, bias mitigation techniques, and complex processing strategies. This technical complexity has limited the adoption of fairness-aware practices in real-world applications.

The research team developed FairAgent, an automated system powered by large language models that democratizes fairness-aware machine learning. FairAgent eliminates the need for specialized expertise by automatically analyzing datasets for potential biases, preprocessing data appropriately, and implementing bias mitigation strategies based on user requirements. The system provides an intuitive web interface where practitioners specify their fairness goals, and FairAgent handles the technical complexities automatically.

The system workflow begins with automated data analysis, where FairAgent examines feature distributions, missing value patterns, and correlations. It then performs contextual analysis to identify sensitive attributes (such as race, gender, or age) and recommends appropriate fairness metrics. After preprocessing the data, FairAgent trains a machine learning model to quantify algorithmic biases. Finally, it builds fairness-aware models using techniques such as adversarial training or constrained learning, with precise hyperparameter tuning to meet user-specified fairness thresholds while maximizing accuracy.

Experiments on benchmark datasets demonstrate that FairAgent achieves fairness thresholds within ±0.005 of target values while maintaining high accuracy. The system supports multiple fairness metrics including demographic parity and equalized odds, and works with various LLM backends including GPT-4o, Claude, and Gemini. By automating complex technical processes and providing accessible workflows, FairAgent represents a significant advance toward making fairness-aware machine learning practical for practitioners without deep technical expertise.