Colloquia - Mengdi Huai, Fostering Trustworthiness in Machine Learning: From Transparence to Robustness, Virtual, 4:25 - 5:25 pm

Friday, December 10, 2021 - 4:25pm to 5:25pm
Event Type: 

Fostering Trustworthiness in Machine Learning: From Transparence to Robustness


Machine learning algorithms have been widely applied in real world to build intelligent systems (e.g., self-driving cars, intelligent recommendation systems, and clinical decision support systems). However, traditional machine learning algorithms mainly focus on optimizing accuracy and efficiency, and they fail to consider how to foster trustworthiness in their design. Trustworthiness reflects the degree of a user's confidence in that the deployed machine learning system will operate as the user expects in the face of various circumstances such as malicious attacks, human errors, and system faults. Without trustworthiness guarantee, the machine learning systems deployed in real world may produce a variety of devastating social and environmental consequences. In this talk, I will introduce my research efforts towards the goal of fostering trustworthiness in machine learning. Specifically, I will focus on two critical trustworthiness issues: transparency and robustness. On the transparency front, I will discuss an important family of learning problems called pairwise learning and present two general interpretation methods that can remove the “black box” and increase the transparency of pairwise models. On the robustness front, I will describe my recent research on the security vulnerability of model interpretation methods for deep reinforcement learning (DRL) and introduce two malicious attack frameworks that can significantly alter the interpretation results while incurring minor damage to the performance of the original DRL model.


Mengdi Huai is a Ph.D. candidate in the Department of Computer Science at the University of Virginia. Her research interests lie in the areas of data mining and machine learning, with a current focus on developing novel techniques to build trustworthy learning systems that are explainable, robust, private, and fair. Mengdi is also interested in designing effective data mining and machine learning algorithms to deal with complex data with both strong empirical performance and theoretical guarantees. Her research work has been published in top-tier data mining and machine learning conferences and journals. She has received multiple research awards, including the Rising Star in EECS at MIT, the Rising Star in Data Science at UChicago, Best Paper Runner-up for KDD2020, the Sture G. Olsson Fellowship in Engineering, and the John A. Stankovic Research Award at the University of Virginia.