The school of Electrical Engineering’s professor Euijong Whang and Changho Suh’s research team has developed a new study method for training artificial intelligence (AI). The research was lead by Ph.D candidate Yuji Roh (PH Euijong Whang) along with professor Kangwook Lee from the Wisconsin Medison department of electrical engineering.
Machine learning provides many convenient and accurate systems for modern society, but there have been many issues about reliability recently. Therefore fairness, robustness, transparency, and explainability have also arisen as an important factor rather than sole accuracy.
The research team has proposed a new method called FR-Train where fairness and robustness can be enhanced within a single framework, which are both deeply related to training data. The FR-Train method is the first framework for fair and robust learning, which can provide an interpretation based on mutual information theory and be used for various applications.
The team announced that the method can be used for the foundation of a fair and robust AI system. Since the issue of reliability will become more and more important in the future, it is expected that many other related researches will be performed. The achievement was presented at ICML (International Conference for Machine Learning) 2020, the most recognized conference in machine learning.
Detailed information can be found in the link below.
Figure 1. FR-Train architecture with loan example
[Paper information and Links]
Title: FR-Train: A Mutual Information-Based Approach to Fair and Robust Training
Authors: Yuji Roh (KAIST EE), Kangwook Lee (Wisconsin-Madison Electrical & Computer Engineering), Steven Euijong Whang (KAIST EE), and Changho Suh (KAIST EE)