Zhifeng Jiang 江志锋｜ Huawei 华为科技有限公司

About Me

I am a Research Scientist at Huawei's Central Software Institute, working on embodied AI systems and on-device AI systems for smart glasses, phones, PCs, and robot prototypes.

I received my PhD in 2024 from Computer Science and Engineering, Hong Kong University of Science and Technology. advised by Prof. Wei Wang. I also work closely with Dr. Ruichuan Chen and Prof. Bo Li. Prior to that, I got my Bachelor in 2019 from Department of Computer Science and Technology, Zhejiang University.

Current Topics: Embodied AI and On-device AI Systems

Embodied AI and Robot Agents:
- VLA-based robot manipulation on SO101 arms.
- Robiagent: A long-horizon robot agent for real-robot manipulation.
First-Person Wearable Agents:
- VisionBot: An AI-glasses-oriented agent for always-on first-person perception and service.
On-device AI Systems:
- LLM inference acceleration for phones.
- Target speech hearing and speech enhancement for AI glasses.
- Hey: Tool-use agents for personal computers.

Selected Projects in Embodied AI and On-device AI

Real Robot RobiAgent: Agentic Dual-Arm Robot Prototype
RobiAgent is a real-robot research prototype for studying long-horizon embodied agents. In a dual-arm showroom scenario, one arm continuously adjusts a smartphone toward the user, while another arm performs touchscreen operations with synchronized speech. I lead a 6-person team to build the system and use it to study real-robot manipulation and real-time, constraint-aware coordination of coupled robot subtasks.
Robot agents · Dual-arm manipulation · Long-horizon tasks
[code]

Real Robot LeRobot-SO101: Real-Robot Imitation Learning Pipeline
An end-to-end real-robot learning pipeline on SO101 arms. I collected demonstrations for a manipulation task where the robot places a banana into a box, trained VLA policies (ACT/SmolVLA/pi0.5), and deployed it back to the physical robot.
LeRobot · SO101 · VLA
[code]

AI Glasses VisionBot: First-Person Agent for AI Glasses
VisionBot is an AI-glasses-oriented agent for always-on first-person perception and service. Unlike chat-style assistants, it is designed to provide continuous service after a one-shot trigger. Current demos include lost-and-found via continuous visual search and fitness coaching via first-person activity logging and feedback.
First-person perception · Wearable agents
[code]

AI Glasses Target Speech Hearing on AI Glasses
A real-time speech enhancement system for noisy face-to-face communication. The system receives stereo audio streams from ordinary omnidirectional microphones on AI glasses, enhances the frontal target speaker, and plays the enhanced speech back through the glasses speaker. I designed the stereo speech enhancement model, built the streaming inference pipeline, and deployed the prototype through a HarmonyOS application.
Speech enhancement · Spatial audio · Streaming inference · HarmonyOS
[Demo available upon request]

Selected Papers

Findings of EMNLP'25 PromptKeeper: Safeguarding System Prompts for LLMs
Zhifeng Jiang, Zhihua Jin, Guoliang He

USENIX Security'24 Lotto: Secure Participant Selection against Adversarial Servers in Federated Learning
Zhifeng Jiang, Peng Ye, Shiqi He, Wei Wang, Ruichuan Chen, Bo Li

ACM EuroSys'24 Dordis: Efficient Federated Learning with Dropout-Resilient Differential Privacy
Zhifeng Jiang, Wei Wang, Ruichuan Chen

IEEE TBD'23 Towards Efficient Synchronous Federated Training: A Survey on System Optimization Strategies
Zhifeng Jiang, Wei Wang, Bo Li, Qiang Yang

ACM SoCC'22 Pisces: Efficient Federated Learning via Guided Asynchronous Training
Zhifeng Jiang, Wei Wang, Baochun Li, Bo Li

Other Papers

TPDS'26 Enabling Client-Autonomous Training Optimizations for Efficient Federated Learning
Na Lv, Jiayi Zhang, Zhi Shen, Chen Chen, Zhifeng Jiang, Quan Chen, Minyi Guo

html

bibtex

IEEE SPL'26 AS-EVNorm: Tail-Aware Extreme Value Normalization for Speaker Verification
Zekai Su, Yongjun He, Jiqing Han, Jianchen Li, Yikun Jiang, Zhifeng Jiang, Tianhong Ding

html

bibtex

IEEE TMC'26 Boosting Gradient-based Training Diagnosis for Efficient and Accurate Federated Learning
Jiayi Zhang, Zuo Gan, Chen Chen, Zhifeng Jiang, Hao Wang, Yifei Zhu, Quan Chen, Minyi Guo

html

bibtex

IEEE TDSC'25 Feature Reconstruction Attacks and Countermeasures of DNN training in Vertical Federated Learning
Peng Ye, Zhifeng Jiang, Wei Wang, Bo Li, Baochun Li

html

code

bibtex

ACM PPoPP'25 SGDRC: Software-Defined Dynamic Resource Control for Concurrent DNN Inference on NVIDIA GPUs
Yongkang Zhang, Haoxuan Yu, Chenxia Han, Cheng Wang, Baotong Lu, Yunzhe Li, Zhifeng Jiang, Yang Li, Xiaowen Chu, Huaicheng Li

html

bibtex

ICPP'24 FedCA: Eﬀicient Federated Learning with Client Autonomy
Na Lv, Zhi Shen, Chen Chen, Zhifeng Jiang, Jiayi Zhang, Quan Chen, Minyi Guo

html

bibtex

IEEE ICDCS'21 Gillis: Serving Large Neural Networks in Serverless Functions with Automatic Model Partitioning
Minchen Yu, Zhifeng Jiang, Hok Chun Ng, Wei Wang, Ruichuan Chen, Bo Li

Best Paper Runner-Up Award