Scholar LinkedIn Github Twitter Resume
Research Scientist at Meta
Bio
Yue (Julien) Niu is a research scientist at Meta, working on training and serving efficiency in large-scale recommendation systems. He obtained his Ph.D. at the University of Southern California (USC). He worked with Professor Salman Avestimehr, the director of Information Theory and Machine Learning (vITAL) Lab. He also once worked with Professor Viktor Prasanna in FPGA/Parallel Computing Lab. Before that, He got his master’s and bachelor’s degree from Northwestern Polytechnical University in Xi’an, China.
His research interests cover various aspects of efficient privacy-preserving machine learning using information theory, differential privacy, federated learning, and distributed setting with private and public environments. He is also conducting research on deep learning and language model acceleration , which reduce training and inference costs using quantization, pruning and low-rank compression.
email: yueniu2022 [at] gmail [dot] com
Research Focus
-
Efficient Private Machine Learning
-
Privacy, bias, and fairness in language models
-
Differentially private machine learning with improved model utility
-
Private machine learning empowered by trusted execution environment
-
-
CNN/Transformer/LLM Acceleration
-
Fast training and inference via low-rank models and low-rank activation
-
Memory-efficient training and inference via low-rank and sparse compression
-
Accelerate neural networks with dedicated hardware
-
-
Federated Learning at the Edge
-
Federated learning of large models at resource-constrained devices
-
Communication-efficient federated learning with sparse training on clients
-
News
2024/6/14: I defend my thesis, Striking the Balance: Optimizing Privacy, Utility and Complexity in Private Machine Learning .
2024/3/13: Our paper, Ethos: Rectifying Language Models in Orthogonal Parameter Space , has been accepted to North American Chapter of the Association for Computational Linguistics (NAACL) Findings, 2024.
2024/2/26: Our paper, All Rivers Run to the Sea: Private Learning with Asymmetric Flows , has been accepted to IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR), 2024.
2024/2/21: I will be attending ITA, 2024 and present our work, All Rivers Run To The Sea: Private Learning with Asymmetric Flows .
2024/1/31: Our paper, Edge Private Graph Neural Networks with Singular Value Perturbation , is accepted to Privacy Enhancing Technologies Symposium (PETs), 2024.
Experience
Research Scientist at Meta
2024/7 - Present
Topic: Training and serving efficiency in modern recommendation systems.
Applied Scientist Intern at Amazon Alexa
2022/6 - 2022/9
Topic: Estimate a CV model’s performance in the wild
Collaborator: Furqan Khan, Pradeep Natarajan, Ruoxi Liu
Applied Scientist Intern at Amazon Alexa
2021/6 - 2021/9
Topic: Personalized model compression using knowledge distillation
Collaborator: Furqan Khan, Pradeep Natarajan, Salman Avestimehr
Research Intern at Tsinghua University
2017/6 - 2018/6
Topic: Neural network acceleration on FPGA
Collaborator: Zhenyu Liu, Xiangyang Ji
Academic Service
Conference Reviewer:
CVPR — 2024(1)
ICLR — 2025(3) — 2024(4) — 2022(2) — 2021(2)
NeurIPS — 2024(6) — 2023(6) — 2022(4)
ICML — 2024(6) — 2023(4)
AAAI — 2025(2)
KDD — 2023(3)
WACV — 2024(2) — 2023(3)
SDM — 2024 (3)
Journal Reviewer:
IEEE Transactions on Machine Learning Research (TMLR) — 2024 (1)
IEEE Transactions on Network Science and Engineering — 2024 (1)
Transactions on Mobile Computing — 2023 (1)
Teaching:
TA for Introduction to Digital Circuits — Spring 2020 — Fall 2019
Awards
Best Poster Award at USC-Amazon Annual Symposium on Secure and Trusted ML
Los Angeles, 2023.