Your Sidebar Name

Your Sidebar Name

Short biography for the left-hand sidebar

Zishuai Zhang’s personal website.

Personal Github

ZishuaiZhang

Contact

Zishuai_Zhang@outlook.com

Education

Master’s Degree

  • Institution: Beijing University of Aeronautics and Astronautics (BUAA)
  • Degree: Master of Science in Artificial Intelligence
  • Duration: September 2023 - Present
  • Research Group: Future Blockchain and Privacy Computing Advanced Research Center, BUAA
  • Research Focus: Federated Splitting Large Models and Privacy Computing
  • Advisor: Professor Hainan Zhang

Bachelor’s Degree

  • Institution: Dalian University of Technology (DUT)
  • Degree: Bachelor of Science in Information and Computing Science (Hua Luogeng Class)
  • Duration: September 2019 - July 2023

Research Experience

Federated Splitting Large Models with Differential Privacy

  • Project: AIJ Paper - “A Federated Splitting Framework for LLMs: Security, Efficiency, and Adaptability” (under review)
  • Outcomes: Developed a framework where large models are split into smaller sub-models distributed across servers and clients, ensuring data privacy while reducing client computational load.

Retrieval-Augmented Generation (RAG)

  • Project: Enhanced large model performance through document retrieval
  • Methodology: Utilized LLamaIndex for vectorized embedding of knowledge bases, enabling similarity matching of user queries with vectorized data to improve model accuracy and reduce computational load.

Blockchain Incentive Mechanism Based on Evolutionary Game Theory

  • Conference Paper: “A Semantic Detection Incentive Mechanism For Blockchain Transactions Based on Evolutionary Game Theory” (EI, IEEE Global Blockchain Conference)
  • Contribution: Proposed a new incentive mechanism for Bitcoin transaction validation, incorporating evolutionary game theory for transaction legality verification and anomaly detection.
  • Patent: “An Incentive Mechanism for Blockchain Transaction Semantic Detection Based on Evolutionary Game Theory” (Pending)

Project Experience

Federated Privacy Large Model at Microchip Research Institute

  • Duration: June 2024 - November 2024
  • Project: Split large models into three parts, with clients holding the first and last Transformer blocks, and servers holding the middle block.
  • Framework: Utilized Flower federated learning framework for data transmission between models.
  • Outcomes: Demonstrated comparable performance between distributed and centralized models, validating the effectiveness of model splitting in protecting privacy and reducing client computational requirements.

Reinforcement Learning for Blockchain Adaptive Optimization at Microchip Research Institute

  • Duration: April 2024 - Present
  • Project: Debugged blockchain transaction pressure measurement and latency impact on blockchain TPS, optimized blockchain parameters using online reinforcement learning (PPO algorithm).
  • Outcomes: Improved blockchain TPS by 8.4% under network latency conditions, proving the effectiveness of the algorithm.

Self-Evaluation

  • Passion: Focused on what he loves, enjoys learning new technologies and applying them in practice, gains a sense of accomplishment from creating value.