CV

B.S. in Information Security

Research on LLM Serving System

• Advisors Prof. Yufei Ding
• We developed an inference method combining speculative models with a tool engine, speculative model engine, and LLM decoding engine to optimize memory allocation and request scheduling, significantly improving the efficiency of CoT and tool-usage LLM inference serving.
• Achieved 30-40% speedup over using only large models by incorporating collaborative reasoning between large and small models on the overall improvement.
• I played a key role in implementing the entire code framework over vLLM, writing over 5k lines of Python code and around 1k lines of CUDA/C++ code, profiling system data, performing evaluations and optimizing inefficient kernel to achieve a 2x speedup on this kernel.
• NOTE: Preparing for submission, SOSP 2025

Research on DLRM System

• Advisors Prof. Yufei Ding
• We proposed a novel solution to optimize DLRM training on disaggregated memory, achieving an average speedup of 5.34× over traditional frameworks leveraging host DRAM and outperformed state-of-the-art GPU-only DLRM training frameworks.
• Our approach leverages CXL technology for flexible memory expansion and introduces an access-aware strategy for embedding placement and communication routing.
• I was responsible for formulating the approach, designing and coding the solver, and running the experiments to validate our solution.
• NOTE: Under Review, ISCA 2025

Research on Robust Image Watermark

• Adivisors Dr. Kejiang Chen
• Proposed a watermark recovery module based on image hiding, designed to restore information-embedded images subjected to cropping and spatial manipulation.
• Completed experiments on the reproduction of related research and developed the code for the module framework as proposed by coworker.

CTF Player

• Member of USTC-NEBULA
• Our team ranks 16th/1655(~0.97%) in China in the CTFTime rating during 2023-2024!
• Mainly focus on Misc (Steganograpy and Forensics)

Tiny-C Compiler

• Developed a simplified C compiler in C, implementing lexical analysis, syntax analysis, and Intermediate Representation (IR) generation for the Tiny-C language.
• Supported key C language features in Tiny-C, including assignment, arithmetic operations, while loops, and if-else statements.
• Github repo Tiny-C Compiler

Semantic Guidance for Diffusion Model

• Conducted a detailed investigation into the semantic guidance of diffusion models, focusing on understanding both the theoretical underpinnings and experimental outcomes.
• Identified existing issues within the semantic guidance provided by diffusion models, particularly emphasizing the loss of positional information.
• Developed a straightforward method to address the loss of positional information in the semantic guidance of diffusion models, enhancing model accuracy and interpretability.
• Blog post Semantic Guidance for Diffusion Model

• Top 5th(~1.5%) in Hackergame2024
• 3rd prize(~20%) in D^3CTF
• Top 9th(~1%) in L3HCTF

• Top 3rd(~1%) in TPCTF
• 3rd prize(~20%) in D^3CTF
• 3rd prize(~6.5%) in Hackergame2023

• Outstanding Student Scholarship
• DAS Scholarship(Top 10 in the department)

• Outstanding Student Scholarship

• Zhang Zongzhi Scholarship

Information Security Design and Practice

• Information hiding in multimedia and file analysis

Data Structure and Algorithm