Zhixiao Wang

Program Details: School of Computer Science, Master of Computer Science and Technology, expected to graduate in June 2028.

Research Interests: Computer Vision; Deep Learning.

Program Details: School of Computer Science and Technology, Bachelor of Engineering in Computer Science and Technology, graduated in June 2025.

Highlights: Core courses in Artificial Intelligence, Operating Systems, and Computer Networks; recognized as Outstanding Graduate.

I led our vision group in developing algorithms for the RMUC, focusing on advanced deep learning models for real-time object recognition and tracking. Our efforts significantly boosted our robots' operational capabilities, leading to our winning the first prize in this prestigious competition that challenges a broad spectrum of engineering skills.

During this period, I self-studied "Dive into Deep Learning" and made use of what I learned by modifying and optimizing the YOLOv5-face model for our RoboMaster competition. This project was crucial for improving the real-time object recognition capabilities of our robots.

I led the team to a championship victory in the BOTEC2022, where we developed advanced visual navigation systems for humanoid robots equipped with infrared structured light depth cameras. The competition focused on autonomous navigation and task execution.

I designed and implemented a pipelined CPU based on the RISC-V instruction set using Verilog. This project enhanced my skills in computer architecture and hardware design. Also my source code is available on GitHub.

I participated in the design and development of a high-performance fault-tolerant storage system based on EVENODD encoding. I contributed to the initial code writing and subsequent algorithm optimization and parallelization, focusing on enhancing system reliability and performance.

I independently completed 11 labs as part of the MIT 6.S081 course, implementing various system-level features within the xv6 operating system. These tasks involved advanced areas such as memory management, concurrency synchronization, and file systems.

As a teaching assistant for the Digital Logic Design course, I was responsible for designing and implementing a new serial communication experiment. This initiative was aimed at enhancing the practical understanding of digital logic among students.

In the ASC24 Supercomputer Competition, I played a key role in optimizing LLM inference and managing power consumption during the preliminary round. Our team's efforts led to high-performance results in the finals, tackling complex tasks in the GoMars and Group Competition events, achieving the first prize.

Led the team in the National Student Computer System Capability Challenge - Operating System Design Contest - OS Function Track. We developed JFUSE (a file system project) focusing on metadata management, caching, and concurrent I/O. Source code is available on GitLab. Awarded Second Prize (National) on Aug 22, 2024.

Efficient Inference for Large Language Models Based on Cache-Enhanced Knowledge. Proposed and built a mentor-mentee dual-model pipeline with a semantic cache (BGE-M3 + FAISS) and prompt-based knowledge augmentation; designed benchmarks and evaluations (e.g., MMLU Pro and math variants); explored GRPO-based fine-tuning to improve cache utilization; reduced redundant computation while maintaining answer quality.