Hi, my name is Mingxuan Song.
I'm currently pursuing my PhD at Peking University, under the supervision of Professor Zhen Xiao.
About me
I am currently a PhD student in Computer Systems Architecture at the School of Computer Science, Peking University, under the supervision of Professor Zhen Xiao, with an expected graduation in July 2028. I received my bachelorβs degree in computer science and technology from China University of Geosciences, Wuhan, in 2023. My research interests include reinforcement learning (RL), sharding blockchain, and large language models (LLMs).
My goal is to continuously grow both professionally and personally, while maintaining a healthy and fulfilling lifestyle. I am also actively seeking internship and job opportunities worldwide. Feel free to contact me!
π Location: Beijing, China
π― Hobbies: πΈ Badminton | π± Billiards | π Table Tennis | πββοΈ Running
View Resumeπ Publications
Scholarπ° News
π Projects
Few-Shot RL Fine-Tuning for LLMs
Affiliations:
School of Computer Science, Peking University;
Alimama, Alibaba Group.
In recent years, Large Language Models (LLMs) have demonstrated remarkable performance across a variety of natural language processing tasks. However, fine-tuning these models typically requires large-scale datasets and extensive computational resources, which limits their applicability in scenarios where data is scarce and budgets are constrained. This work explores a novel approach to few-shot reinforcement learning (RL) fine-tuning for LLMs, aiming to adapt pre-trained models to specific tasks using minimal supervision.
RL for Efficient Sharding Blockchain
Affiliations:
School of Computer Science, Peking University;
Theta Labs, Theta Inc.
Sharding blockchain systems face critical challenges in achieving efficient cross-shard data distribution and maintaining balanced workload across shards. Traditional address allocation methods often suffer from high latency and uneven shard utilization, especially when dealing with dynamically changing transaction patterns and reconfiguration events.
RL for Advanced NFT Auction System
Affiliations:
School of Computer Science, Peking University;
Theta Labs, Theta Inc.
In the Web 3.0 era, NFTs have become a popular asset type, and the auction system is a crucial component of the NFT ecosystem. However, traditional auction systems often suffer from low throughput and high latency, especially when dealing with high-demand auctions. To address these challenges, we propose a hierarchical reinforcement learning approach to optimize dynamic bidding strategies, enabling more efficient and responsive NFT auctions.