BIO

I am a third-year PhD student in the School of Computer Science at Carnegie Mellon University(CMU) and Student Researcher at Google DeepMind. My research focuses on Large Language Model post-training, make the LLMs faster (more efficient in training and inference), and cheaper (training with less GPUs/GPU hours and cheaper to deployment), and better (more align with domain specific tasks and users).

At CMU, I am advised by Prof. Heather Miller. Previously, I earned my master in Computer Science from New York University and my master thesis is supervised by Prof. Anna Choromanska. I received my B.S. in Computer Science and Engineering from The Chinese University of Hong Kong(CUHK). Before starting my PhD, My research mainly focuses on distributed machine learning system.

• (This personal website is updated as of July 2026.)

News

• 2026: I start my internship at Google DeepMind, hosted by Samira Khan and Majid Hadian (Gemini Long Context) and Yu Gan and Arvind Krishnamurthy (SRG).
• 2026: Less is MoE: Trimming Experts in Domain-Specialist Language Models has been accepted by EMNLP 2026 (Main Conference)!
• 2026: Preserving Long-Tailed Expert Information in Mixture-of-Experts Tuning has been accepted by COLM 2026! We open-sourced the code.
• 2026: RAST-MoE-RL: A Regime-Aware Spatio-Temporal MoE Framework for Deep Reinforcement Learning in Ride-Hailing has been accepted by ICML 2026.
• 2025: I intern at AWS-AI-Labs@Amazon this summer working on LLM post-training enabled speculative decoding via latent space reasoning.
• 2025: Open-source SMT. We implemented SMT in two frameworks: DeepSpeed and Hugging Face Trainer.
• 2025: SMT: Fine-Tuning Large Language Models with Sparse Matrices has been accepted by ICLR 2025.
• 2024: Adjacent Leader Decentralized Stochastic Gradient Descent has been accepted by ECAI 2024.
• 2024: Multi-View Radar Autoencoder for Self-Supervised Automotive Radar Representation Learning has been accepted by IEEE Intelligent Vehicles Symposium (IV) 2024.
• 2023: I started my Ph.D. journey at CMU.

Selected Publications

• Haoze He*, Xinkai Zou*, Xuan Jiang, Xingyuan Ding, Ao Qu, Juncheng Billy Li, Heather Miller, “Less is MoE: Trimming Experts in Domain-Specialist Language Models”, Conference on Empirical Methods in Natural Language Processing (EMNLP), Main Conference, Accepted, 2026.

• Haoze He, Xingyuan Ding, Xuan Jiang, Alex Cheng, Yibo Zhao, Juncheng Billy Li+, Heather Miller+, “Preserving Long-Tailed Expert Information in Mixture-of-Experts Tuning”, Conference on Language Modeling (COLM), Published, 2026. [code]

• Yuhan Tang*, Kangxin Cui*, Jung Ho Park*, Yibo Zhao*, Xuan Jiang+, Haoze He+, Jiangbo Yu, Haris Koutsopoulos, Jinhua Zhao, “RAST-MoE-RL: A Regime-Aware Spatio-Temporal MoE Framework for Deep Reinforcement Learning in Ride-Hailing”, International Conference on Machine Learning (ICML), Accepted, 2026.

• Haoze He*, Juncheng Billy Li*, Xuan Jiang, Heather Miller, “Sparse Matrix in Large Language Model Fine-Tuning”, International Conference on Learning Representations (ICLR), Accepted, Jan. 2025. [code]

• Haoze He*, Jing Wang*, Anna Choromanska, “Adjacent Leader Decentralized Stochastic Gradient Descent”, European Conference on Artificial Intelligence (ECAI), Accepted, June 2024. [code]

My full publication list can be found on my Google Scholar profile. (*: equal contribution; +: corresponding author)

Academic Blog

• Peter Zhong, Haoze He, Omar Khattab, Christopher Potts, Matei Zaharia, Heather Miller, “A Guide to Large Language Model Abstractions”, Jan. 2024.

Education

• Ph.D. in Machine Learning and Software Engineering at Carnegie Mellon University, 2023-present
  • GPA: 4.16/4.0, Rank: top1%
• M.S. in Computer Engineering at New York University, 2021-2023
  • GPA: 3.93/4.0, Rank: top1%
• B.S. in Computer Science and Engineering at The Chinese University of Hong Kong, 2016-2020