Wenzhi Fang
Wenzhi Fang

PhD student

Purdue University
Hello and welcome! I am currently a third-year PhD student at Purdue University, majoring in Electrical and Computer Engineering, advised by Prof. Christopher G. Brinton. Previous to that, I obtained my master’s degree in Electrical and Computer Engineering at ShanghaiTech University under the supervision of Prof. Yong Zhou and Prof. Yuanming Shi. In addition, from Aug. 2022 to Feb. 2023, I served as a research intern in the Optimization for Machine Learning lab at KAUST led by Prof. Peter Richtárik.
Download CV
Interests
  • RL-based Post-training
  • Efficient Fine-Tuning of LLM
  • Federated Learning
Education

  • PhD Electrical and Computer Engineering

    Purdue University

  • MS Electrical and Computer Engineering

    ShanghaiTech

  • BS Electrical Engineering

    Shanghai University

Research Pulse

My work focuses on LLM post-training and Multi-agent LLM systems:

  • RL-based Post Training, developing reinforcement learning frameworks for adaptive control and collaboration of multi-agent LLM systems.
  • Efficient On-deive LLM fine-tuning, enabling distributed on-device LLM fine-tuning under computation, communication, and memory constraints.
  • Federated learning and optimization, designing efficient and convergent algroithm for distributed model training.
Experience & training

PhD, Purdue University — Electrical & Computer Engineering (Aug 2023 — Present)
Working on LLM post-training and reasoning, with an emphasis on RL-based alignment, agent collaboration, and communication-efficient fine-tuning for distributed on-device LLM systems.

Research Intern, Optimization for Machine Learning Lab, KAUST (Aug 2022 — Feb 2023)
Working on optimization theory and its applications on machine learning, especially primal-dual algorithm.

MS, ShanghaiTech University — Electrical & Computer Engineering (2020 — 2023)
Working on optimization theory and its applications on communication system.

News
  • Feb 2026: Ou work on DA-GRPO, a budget-aware reinforcement learning framework enabling continual learning in device–cloud LLM systems is now on ArXiv.
  • Sept 2025: Our work on device–cloud collaborative LLM reasoning, introducing an RL-based unified framework for adaptive routing and post-training is now on ArXiv.
  • Jan 2025: Our work on Federated Sketching LoRA (FSLoRA), a communication-efficient framework for collaborative on-device LLM fine-tuning under heterogeneous resource constraints is now on ArXiv.
  • Sept 2024: Our paper on hierarchical federated learning with multi-timescale gradient correction (MTGC) has been accepted to NeurIPS 2024.
  • Aug 2023: Joined Purdue as a PhD student after completing my MS at ShanghaiTech.
Selected works
Joint Continual Learning of Local Language Models and Cloud Offloading Decisions with Budget Constraints

Joint Continual Learning of Local Language Models and Cloud Offloading Decisions with Budget Constraints

Continual post-training framework (DA-GRPO) that jointly learns on-device SLM policies and budget-aware cloud offloading decisions to reduce forgetting while respecting assistance constraints.

Evan Chen*, Wenzhi Fang*, Shiqiang Wang, Christopher Brinton
arXiv preprint · January 2026
Bridging On-Device and Cloud LLMs for Collaborative Reasoning: A Unified Methodology for Local Routing and Post-Training

Bridging On-Device and Cloud LLMs for Collaborative Reasoning: A Unified Methodology for Local Routing and Post-Training

RL-based framework that enables on-device LLMs to decide when to invoke cloud models, jointly learning routing and post-training to balance accuracy and compute.

Wenzhi Fang, Dong-Jun Han, Liangqi Yuan, Evan Chen, Christopher G. Brinton
arXiv preprint · September 2025
TAP: Two-Stage Adaptive Personalization of Multi-task and Multi-Modal Foundation Models in Federated Learning

TAP: Two-Stage Adaptive Personalization of Multi-task and Multi-Modal Foundation Models in Federated Learning

Personalization algorithm for multi-task, multi-modal foundation models that combines architecture-aware replacement with post-FL knowledge distillation.

Seohyun Lee, Wenzhi Fang, Dong-Jun Han, Seyyedali Hosseinalipour, Christopher G. Brinton
arXiv preprint · September 2025
Federated Sketching LoRA: On-Device Collaborative Fine-Tuning of Large Language Models

Federated Sketching LoRA: On-Device Collaborative Fine-Tuning of Large Language Models

Communication-efficient framework that uses sketching to adapt LoRA ranks per device, enabling collaborative on-device LLM fine-tuning under heterogeneous resources.

Wenzhi Fang, Dong-Jun Han, Liangqi Yuan, Seyyedali Hosseinalipour, Christopher G. Brinton
arXiv preprint · January 2025
Hierarchical Federated Learning with Multi-Timescale Gradient Correction

Hierarchical Federated Learning with Multi-Timescale Gradient Correction

Multi-timescale gradient correction method that stabilizes hierarchical FL across client and group levels, with convergence guarantees robust to data heterogeneity.

Wenzhi Fang, Dong-Jun Han, Evan Chen, Shiqiang Wang, Christopher G. Brinton
NeurIPS 2024 · September 2024
Federated Learning over Hierarchical Wireless Networks: Training Latency Minimization via Submodel Partitioning

Federated Learning over Hierarchical Wireless Networks: Training Latency Minimization via Submodel Partitioning

HIST framework that partitions large models into submodels across cells to reduce computation, storage, and communication while meeting convergence guarantees.

Wenzhi Fang, Dong-Jun Han, Christopher G. Brinton
IEEE/ACM Transactions on Networking · February 2024
Communication-efficient stochastic zeroth-order optimization for federated learning

Communication-efficient stochastic zeroth-order optimization for federated learning

FedZO algorithm for derivative-free federated optimization, with AirComp-assisted variant to preserve convergence under noisy wireless aggregation.

Wenzhi Fang, Ziyi Yu, Yuning Jiang, Yuanming Shi, Colin N. Jones, Yong Zhou
IEEE Transactions on Signal Processing · January 2022
Academic Services

Conference Reviewer

Journal Reviewer