CIS Colloquium - by Prof. Meisam Razaviyayn
Event details
| Date | 29.09.2023 |
| Hour | 11:15 › 12:15 |
| Speaker | Prof. Meisam Razaviyayn |
| Location | Online |
| Category | Conferences - Seminars |
| Event Language | English |
Title: A scalable stochastic optimization framework for robust and private fair learning
Abstract: Machine learning models are increasingly used in high-stakes decision-making systems. In such applications, a major concern is that these models sometimes discriminate against certain demographic groups such as individuals with certain race, gender, or age. Another major concern in these applications is the violation of the privacy of users. While fair learning algorithms have been developed to mitigate discrimination issues, these algorithms can still leak sensitive information, such as individuals’ health or financial records. Utilizing the notion of differential privacy (DP), prior works aimed at developing learning algorithms that are both private and fair. However, existing algorithms for DP fair learning require a full-batch of data in each iteration of the algorithm to be able to impose fairness. Moreover, the fairness/accuracy of the model can degrade significantly in prior DP training algorithms. In this work, we developed a min-batch (stochastic) differentially private algorithm for fair learning (with a theoretical convergence guarantee). Here, the term “stochastic” refers to the fact that our proposed algorithm converges even when mini-batches of data are used at each iteration. Our framework is flexible enough to permit different fairness notions, including demographic parity and equalized odds. In addition, our algorithm can be extended to handle scenarios where there is a distribution shift between the train and test data by utilizing a novel distributional robust optimization (DRO) framework. Our numerical experiments show that the proposed algorithm consistently offers significant performance gains over the state-of-the-art baselines, and can be applied to larger-scale problems with non-binary target/sensitive attributes. This is a joint work with Andrew Lowy (UW-Madison), Sina Baharlouei (USC), Devansh Gupta (USC), Rakesh Pavan (UW), Maher Nouiehed (AUB), and Ahmad Beirami (Google Research).
Bio: https://sites.usc.edu/razaviyayn
Meisam Razaviyayn is an associate professor of Industrial and Systems Engineering, Computer Science, Quantitative and Computational Biology, and Electrical Engineering at the University of Southern California. He is also the associate director of the USC-Meta Center for Research and Education in AI and Learning (https://realai.usc.edu). Prior to joining USC, he was a postdoctoral research fellow in the Department of Electrical Engineering at Stanford University. He received his PhD in Electrical Engineering with a minor in Computer Science at the University of Minnesota. He obtained his M.Sc. degree in Mathematics from the University of Minnesota. Meisam Razaviyayn is the recipient of the 2022 NSF CAREER Award, the 2022 Northrop Grumman Excellence in Teaching Award, the 2021 AFOSR Young Investigator Award, the 2021 3M Nontenured Faculty Award, the 2020 ICCM Best Paper Award in Mathematics, IEEE Data Science Workshop Best Paper Award in 2019, the Signal Processing Society Young Author Best Paper Award in 2014, and the finalist for Best Paper Prize for Young Researcher in Continuous Optimization in 2013 and 2016. He is also the silver medalist of Iran’s National Mathematics Olympiad. His research interests include the design and the study of the fundamental aspects of optimization algorithms that arise in the modern data science era.
The Center for Intelligent Systems at EPFL (CIS) is a collaboration among IC, ENAC, SB; SV and STI that brings together researchers working on different aspects of Intelligent Systems. In June 2020, CIS has launched its CIS Colloquia featuring invited notable speakers.
More info
Open to all – Hybrid or on-site BM 52 02 - Zoom link
Abstract: Machine learning models are increasingly used in high-stakes decision-making systems. In such applications, a major concern is that these models sometimes discriminate against certain demographic groups such as individuals with certain race, gender, or age. Another major concern in these applications is the violation of the privacy of users. While fair learning algorithms have been developed to mitigate discrimination issues, these algorithms can still leak sensitive information, such as individuals’ health or financial records. Utilizing the notion of differential privacy (DP), prior works aimed at developing learning algorithms that are both private and fair. However, existing algorithms for DP fair learning require a full-batch of data in each iteration of the algorithm to be able to impose fairness. Moreover, the fairness/accuracy of the model can degrade significantly in prior DP training algorithms. In this work, we developed a min-batch (stochastic) differentially private algorithm for fair learning (with a theoretical convergence guarantee). Here, the term “stochastic” refers to the fact that our proposed algorithm converges even when mini-batches of data are used at each iteration. Our framework is flexible enough to permit different fairness notions, including demographic parity and equalized odds. In addition, our algorithm can be extended to handle scenarios where there is a distribution shift between the train and test data by utilizing a novel distributional robust optimization (DRO) framework. Our numerical experiments show that the proposed algorithm consistently offers significant performance gains over the state-of-the-art baselines, and can be applied to larger-scale problems with non-binary target/sensitive attributes. This is a joint work with Andrew Lowy (UW-Madison), Sina Baharlouei (USC), Devansh Gupta (USC), Rakesh Pavan (UW), Maher Nouiehed (AUB), and Ahmad Beirami (Google Research).
Bio: https://sites.usc.edu/razaviyayn
Meisam Razaviyayn is an associate professor of Industrial and Systems Engineering, Computer Science, Quantitative and Computational Biology, and Electrical Engineering at the University of Southern California. He is also the associate director of the USC-Meta Center for Research and Education in AI and Learning (https://realai.usc.edu). Prior to joining USC, he was a postdoctoral research fellow in the Department of Electrical Engineering at Stanford University. He received his PhD in Electrical Engineering with a minor in Computer Science at the University of Minnesota. He obtained his M.Sc. degree in Mathematics from the University of Minnesota. Meisam Razaviyayn is the recipient of the 2022 NSF CAREER Award, the 2022 Northrop Grumman Excellence in Teaching Award, the 2021 AFOSR Young Investigator Award, the 2021 3M Nontenured Faculty Award, the 2020 ICCM Best Paper Award in Mathematics, IEEE Data Science Workshop Best Paper Award in 2019, the Signal Processing Society Young Author Best Paper Award in 2014, and the finalist for Best Paper Prize for Young Researcher in Continuous Optimization in 2013 and 2016. He is also the silver medalist of Iran’s National Mathematics Olympiad. His research interests include the design and the study of the fundamental aspects of optimization algorithms that arise in the modern data science era.
The Center for Intelligent Systems at EPFL (CIS) is a collaboration among IC, ENAC, SB; SV and STI that brings together researchers working on different aspects of Intelligent Systems. In June 2020, CIS has launched its CIS Colloquia featuring invited notable speakers.
More info
Open to all – Hybrid or on-site BM 52 02 - Zoom link
Links
Practical information
- General public
- Free
Organizer
- CIS