I am a fifth-year PhD student in Machine Learning Department at Carnegie Mellon University advised by Virginia Smith. Prior to that, I did my BA at Cornell University double major in Computer Science and Mathematics.

My research interest broadly lies in LLM alignment, machine unlearning, differential privacy, and trustworthy federated learning methods. My works focus on studying analytical formulations and efficient methods that provide both theoretical guarantees and strong empirical performance on real world applications to handle these issues in federated learning.

Research

Safety in LLM

Position: LLM Unlearning Benchmarks are Weak Measures of Progress
Pratiksha Thaker, Shengyuan Hu, Neil Kale, Yash Maurya, Zhiwei Steven Wu, Virginia Smith
Preprint
[PDF]

Jogging the Memory of Unlearned Model Through Targeted Relearning Attack
Shengyuan Hu, Yiwei Fu, Zhiwei Steven Wu, Virginia Smith
Preprint
[PDF]

Guardrail Baselines for Unlearning in LLMs
Pratiksha Thaker, Yash Maurya, Shengyuan Hu, Zhiwei Steven Wu, Virginia Smith
Preprint
[PDF]

No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices
Qi Pang, Shengyuan Hu, Wenting Zheng, Virginia Smith
NeurIPS 2024
[PDF][Blog post]

Privacy

Privacy Amplification for the Gaussian Mechanism via Bounded Support
Shengyuan Hu, Saeed Mahloujifar, Virginia Smith, Kamalika Chaudhuri, Chuan Gup
Preprint
[PDF]

On Privacy and Personalization in Cross-Silo Federated Learning
Ziyu Liu, Shengyuan Hu, Steven Wu, Virginia Smith
NeurIPS 2022
[PDF] [Code]

Private Multi-Task Learning: Formulation and Applications to Federated Learning
Shengyuan Hu, Steven Wu, Virginia Smith
TMLR
[PDF]

Trustworthy Federated Learning

Fair Federated Learning via Bounded Group Loss
Shengyuan Hu, Steven Wu, Virginia Smith
SaTML 2023
Best Paper Award at ICLR 2022 Workshop on Socially Responsible Machine Learning (SRML)
[PDF]

Federated Learning as Network Effect Game
Dung Daniel Ngo, Shengyuan Hu, Shuran Zheng, Virginia Smith, Steven Wu
Preprint
[PDF]

FedSynth: Gradient Compression via Synthetic Data in Federated Learning
Shengyuan Hu, Jack Goetz, Kshitiz Malik, Hongyuan Zhan, Zhe Liu, Yue Liu
Presented at FL-NeurIPS’22 Workshop
[PDF]

Ditto: Fair and Robust Federated Learning Through Personalization
Tian Li, Shengyuan Hu, Ahmad Beirami, Virginia Smith
ICML 2021
Best Paper Award at ICLR 2021 Secure ML Workshop
[PDF] [Code(TensorFlow)] [Code(PyTorch)]

Adversarial Robustness

A New Defense Against Adversarial Images: Turning a Weakness into a Strength
Shengyuan Hu*, Tao Yu*, Chuan Guo, Wei-Lun Chao, Kilian Q. Weinberger
NeurIPS 2020
[PDF] [Code]

Others

Selection pressure on the rhizosphere microbiome can alter nitrogen use efficiency and seed yield in Brassica rapa
Joshua Garcia, Joshua Garcia, LiPing Wei, Liang Cheng, Shengyuan Hu, Jed Sparks, James Giovannoni, Jenny Kao-Kniffin
Nature Communications Biology
[PDF]

Professional Service

• Reviewer for ICML (2021-2024)
• Reviewer for ICLR (2021-2024)
• Reviewer For NeurIPS (2021-2024)
• Reviewer for TMLR
• PC Member for NeurIPS 2022 Workshop on Trustworthy and Socially Responsible Machine Learning (TSRML)
• PC Member for FL4NLP@ACL 2022 Workshop
• Reviewer for IEEE Network Magazine

Industrial Experience

Research Intern (2023)
FAIR
San Francisco SF

Research Intern (2021)
Meta Platform, Inc.
Pittsburgh PA (Remote)

Research Enigneer Intern (2019)
Uber Technologies, Inc.
Pittsburgh, PA

Software Engineer Intern (2018)
MasterClass
San Francisco, CA

Teaching Experience

I have served as a Teaching Assistant at Cornell and CMU for the following courses:

• 10-714 Deep Learning System (CMU, 2022)
• 10-605 ML with Large Datasets (CMU, 2021)
• CS 4780 Intrdocution To Machine Learning (Cornell, 2019)
• CS 4820 Introduction to Analysis of Algorithms (Cornell, 2017-2018)
• CS 2800 Discrete Structures (Cornell, 2017)