Thomas Goldstein and Furong Huang Receive 2020 JPMorgan Faculty Research Award

 

Associate Perotto Professor Thomas Goldstein and Assistant Professor Furong Huang recently received the 2020 JPMorgan Faculty Research Award for their project titled “Robust, Private and Fair ML for Financial Models”.

The focus of the project is towards security vulnerabilities in the Financial machine learning models which include the adversarial attacks, privacy leaks, and fairness violations.

The research has three-fold goals:

·  Studying how adversarial attacks impact markets

·  Preventing leaks of the user’s private data and

·  Safeguarding the protected attributes of the customers in order to make fair decisions.

“Our project is directed towards creating a robust toolbox for industry practitioners to analyze their models for security threats, safeguard their users sensitive information from dangerous leaks, and have confidence that models in production satisfy rigorous fairness standards” said Goldstein.

While Goldstein research lies at the intersection of security threats to the AI systems,  machine learning and optimization, and targets applications in computer vision and signal processing, Huang  is an expert on high dimensional statistics, (differentially private) generative models, and understanding uncertainty, generalization and robustness in neural networks through tensor methods.

“Our research concerns the ethical aspects of the financial models developed through AI used in our everyday life. Our effort to make financial models accountable will protect the markets from adversarial attacks, prevent leakage of user privacy, and ensure fairness regarding decisions toward such as loan approval” said Huang. 

The outcomes of the research work will be distributed freely to the public via open publication sources.

In addition to the Department of Computer Science, both Goldstein and Huang hold joint appointments in UMIACS.

More about the Goldstein’s Research- Here

More about Huang’s Research - Here

 

The Department welcomes comments, suggestions and corrections.  Send email to editor [-at-] cs [dot] umd [dot] edu.