Andrew Lowy

I am a postdoctoral Research Associate at University of Wisconsin-Madison,  advised by Stephen J. Wright and working closely with Jonathan UllmanI received my PhD in Applied Math at University of Southern California, advised by Meisam Razaviyayn.  I was awarded the 2023 Center for Applied Mathematical Sciences (CAMS) Graduate Student  Prize for excellence in research with a substantial mathematical component.  Prior to attending USC, I received my B.A. in Public Policy at Princeton University, worked in industry, and completed a post-bac in Mathematics at Columbia University. 

My research interests lie in trustworthy machine learning and optimization, with a focus on privacy, fairness, and robustness. My main area of expertise is differentially private optimization for machine learning.  I especially enjoy understanding fundamental limits and developing scalable algorithms that attain these limits.


Preprints


A Stochastic Optimization Framework for Private and Fair Learning from Decentralized Learning [arXiv]

Devansh Gupta,  A.S. Poornash, Andrew Lowy, Meisam Razaviyayn 

Under review 



Publications 


Optimal Rates for Robust Stochastic Convex Optimization [arXiv]

Changyu Gao, Andrew Lowy, Xingyu Zhao, Stephen J. Wright

Symposium on the Foundations of Responsible Computing, 2025


Faster Algorithms for User-Level Differentially Private Stochastic Convex Optimization [arXiv]

Andrew Lowy*, Daogao Liu*, Hilal Asi* (*reverse alphabetical)

Conference on Neural Information Processing Systems (NeurIPS) 2024 


Analyzing Inference Privacy Risks Through Gradients In Machine Learning [arXiv]

Zhuohang Li,  Andrew Lowy,  Jing Liu, Toshiaki Koike-Akino, Kieran Parsons, Ye Wang

ACM Conference on Computer and Communications Security (CCS) 2024 


Efficient Differentially Private Fine-Tuning of Diffusion Models [arXiv]

Jing Liu, Andrew Lowy, Toshiaki Koike-Akino, Kieran Parsons, Ye Wang

ICML 2024 Next GenAI Safety Workshop 


How to Make the Gradients Small Privately: Improved Rates for Differentially Private Non-Convex Optimization [arXiv; Code]

Andrew Lowy, Jonathan Ullman, Stephen J. Wright

International Conference on Machine Learning (ICML) 2024 

Theory and Practice of Differential Privacy (TPDP) 2024


Private Heterogeneous Federated Learning Without a Trusted Server Revisited: Error-Optimal and Communication-Efficient Algorithms for Convex Losses [arXiv]

Changyu Gao*, Andrew Lowy*, Xingyu Zhao*, Stephen J. Wright (*alphabetical order)

International Conference on Machine Learning (ICML) 2024 

Theory and Practice of Differential Privacy (TPDP) 2024


Optimal Differentially Private Model Training with Public Data [arXiv; Talk; Code]

Andrew Lowy, Zeman Li, Tianjian Huang, and Meisam Razaviyayn

International Conference on Machine Learning (ICML) 2024 

Theory and Practice of Differential Privacy (TPDP) 2023


Why Does Differential Privacy with Large Epsilon Prevent Practical Membership Inference Attacks?  [arXiv]

Andrew Lowy, Zhuohang Li, Jing Liu, Toshiaki Koike-Akino, Kieran Parsons, Ye Wang

AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-24) 


Exploring User-Level Gradient Inversion with a Diffusion Prior [arXiv]

Zhuohang Li, Andrew Lowy, Jing Liu, Toshiaki Koike-Akino, Bradley Malin, Kieran Parsons, Ye Wang

International Workshop on Federated Learning in the Age of Foundation Models at NeurIPS 2023 


Private Stochastic Optimization With Large Worst-Case Lipschitz Parameter [arXiv; Talk (starts at 21:45)] 

Andrew Lowy and Meisam Razaviyayn 

Journal of Privacy and Confidentiality (TPDP 2023 Special Issue selection, to appear)

International Conference on Algorithmic Learning Theory (ALT) 2023

Theory and Practice of Differential Privacy (TPDP) 2023


Stochastic Differentially Private and Fair Learning [arXiv; Talk; Code] 

Andrew Lowy, Devansh Gupta, and Meisam Razaviyayn 

International Conference on Learning Representations (ICLR) 2023

Oral (top 6 papers) in AFCP 2022 at NeurIPS, PMLR 2022

Theory and Practice of Differential Privacy (TPDP) 2023


Private Non-Convex Federated Learning Without a Trusted Server [arXiv; Talk; Code]

Andrew Lowy, Ali Ghafelebashi, and Meisam Razaviyayn 

International Conference on Artificial Intelligence and Statistics (AISTATS) 2023  

Theory and Practice of Differential Privacy (TPDP) 2022


Private Federated Learning Without a Trusted Server:  Optimal Algorithms for Convex Losses [arXiv; Code]

Andrew Lowy and Meisam Razaviyayn 

International Conference on Learning Representations (ICLR) 2023

Theory and Practice of Differential Privacy (TPDP) 2022


A Stochastic Optimization Framework for Fair Risk Minimization [arXiv; Talk; Code] 

Andrew Lowy*, Sina Baharlouei*, Rakesh Pavan, Meisam Razaviyayn, and Ahmad Beirami (*= equal contribution)

Transactions on Machine Learning Research (TMLR), 2022 

Contributed talk in  Trustworthy and Socially Responsible ML (TSRML) at NeurIPS 2022


Efficient Search of First-Order Nash Equilibria in Nonconvex-Concave Smooth Min-Max Problems [arXiv] 

Dmitrii Ostrovskii, Andrew Lowy, and Meisam Razaviyayn 

SIAM Journal of Optimization, 2021 


Output Perturbation for Differentially Private Convex Optimization [arXiv] 

Andrew Lowy and Meisam Razaviyayn 

Spotlight in AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-21) 


Recent Presentations 

October 2024: INFORMS Annual Conference, Seattle, WA

August 2024: 2 Contributed Posters at TPDP 2024, Boston, MA

July 2024: 4 Contributed Posters at ICML 2024, Vienna, Austria

April 2024: IFDS IDEA Seminar, University of Wisconsin-Madison

November 2023: IFDS IDEA Seminar, University of Wisconsin-Madison

September 2023: 3 contributed posters at TPDP, Boston University

June 2023: Invited Talk at USDA's NASS ML Speaker Series, Virtual

April 2023: Invited Talk at Probability & Statistics Seminar, USC 

March 2023: Invited Talk, Harvard University (Virtual)  

Feb. 2023: Contributed Talk at ALT 2023, National University of Singapore, Singapore

Dec. 2022: 3 contributed presentations at NeurIPS 2022 (including one oral), New Orleans, LA

Oct. 2022: Invited Talk at USC-META REAL First Anniversary Event, USC

Sept. 2022: Invited Talk at MLOPT Idea Seminar, UW-Madison (Virtual)

July 2022: Invited Talk at ICCOPT, Lehigh University

July 2022: 2 contributed presentations at TPDP at ICML 2022, Baltimore, MD

March 2022: Invited Talk at USC-META REAL Workshop, USC

Mar. 2022: Invited Talk at Informs Optimization Society, Clemson University

Dec. 2021:  New Frontiers in Federated Learning at NeurIPS 2021, Virtual

July 2021: Contributed Talk at Socially Responsible Machine Learning, ICML 2021, Virtual

Feb. 2021: Spotlight Talk at AAAI Workshop on Privacy-Preserving AI, Virtual

Nov. 2020. Invited Talk at Informs Annual Meeting, Virtual


Teaching Assistantships

Mathematical Statistics (Math 408 at USC, Fall 2021 with Steve Heilman)

Probability Theory (Math 407, Fall 2020 with Steve Heilman)

 Calculus III (Math 226, Spring 2020 with Michael Hall) 

Introductory Statistics (Math 114, Spring 2019 with Cindy Blois)

Differential Equations and Linear Algebra (Math 225, Fall 2019 with David Crombeque)

Calculus II (Math 126, Spring 2018/Fall 2018 with Nabil Ziane/Cymra Haskell)


Professional Service

Conference Reviewer: ICML (2022 Outstanding Reviewer Award), NeurIPS,  COLT, ICLR, AISTATS

Program Commitee Membership: Theory & Practice of Differential Privacy (TPDP) 2024

Journal Reviewer: TMLR (2022 - Present)

Session Chair/Organizer: ICML 2022; Informs Optimization Society (IOS 2022)

Seminar Organizer: IFDS Ideas Forum at UW-Madison (2024-)