Haodong Lei

Haodong Lei

Enjoy your life and work. Gods and ghosts, the Great Wall, islands, oceans, volcanoes, lion heads, snow-capped mountains, rivers and cherry blossoms. These are what I've experienced.

About Me

I am currently pursuing a Master of Engineering in Computer Science and Technology at the PALM Laboratory, School of Software, Southeast University (SEU), Nanjing, China. I enrolled in this program in 2024, driven by my passion for advancing my expertise in computer science and artificial intelligence. My tutor is Hongsong Wang.

Email: l870572761@gmail.com

WeChat: lighting870

Education

2024 - Present: Southeast University (SEU)

2020 - 2024: Hainan University (HNU)

Publications

  1. PEANUT: Fast Inference of Visual Autoregressive Model with Adjacency-Adaptive Dynamical Draft Trees (ACMMM25 Submitted, First Author)
  2. EasyTune: Efficient Step-Aware Fine-Tuning for Diffusion-Based Motion Generation (NeurIPS25 Submitted, Second Author)
  3. MAS-UNet: A U-shaped Network for Prostate Segmentation (Frontiers in Medicine, SCI Q3)

Blogs

  1. 研究Agent的第一篇：CoT、ToT、GoT\&MLLM CoT
  2. Why is Deepseek? 研究DeepSeek的第一篇

Current Research

Speculative Sampling (2024/9 ~ Present)

I am currently focusing on Speculative Sampling, a novel technique designed to accelerate the decoding process in large language models. This method leverages a smaller, faster draft model to generate multiple tokens simultaneously, which are then verified and refined by a larger target model in a single pass. By aligning the output distribution with that of the target model, Speculative Sampling significantly improves computational efficiency without sacrificing accuracy, making it a promising approach for optimizing real-time AI applications, such as natural language generation.

Research Experience

During 2024, I conducted research on AI Agents, exploring how artificial intelligence can be leveraged to create systems capable of autonomous decision-making and task execution. This experience ignited my interest in AI and its practical applications. Currently, as a Master’s student, I am focusing on Speculative Sampling, a cutting-edge technique aimed at enhancing the efficiency and accuracy of machine learning models, particularly in large language models. My work at the PALM Laboratory allows me to contribute to innovative solutions in pattern learning and data mining.