CV

Yiding Wang

Beijing, China
Blancokdb@gmail | 136 8316 9535 | yiding-w.github.io

Education

Peking University, Tong Class, Yuanpei College
Sept 2022 – Present

• Major: Artificial Intelligence (AI)

• GPA: 3.795/4.0

  89.2/100

• Core Courses: Machine Learning (93.5), Mathematical Foundation for AI (93), Multi-agent System (91), Data Structure and Algorithm (91), Probability Theory and Statistics (91)

Research Experience

LLM-based Agents: Desire-driven Autonomy

Peking University, Beijing Institute for General Artificial Intelligence (BIGAI)
Mar 2024 – Sept 2024

• Designed and developed Desire-driven Autonomy (D2A) to simulate human-like daily activities
• Demonstrated that the Desire-driven autonomous agent significantly outperforms baseline agents in generating realistic activity sequences, as evaluated by GPT-4 and human annotators
• Authored a paper based on this work, which is currently under submission

Parameter-Efficient Fine-Tuning (PEFT)

Peking University
Oct 2024 – Present

• Designed a distributed LLMs fine-tuning framework, where model parameters along different directions are fine-tuned across multiple GPUs
• Developed a synchronization mechanism to ensure consistency of aggregated model parameters across GPUs

Large Language Models Reasoning

Peking University
Apr 2024 – Sept 2024

• Explored using LLMs to automatically label known information and prior reasoning steps, dynamically selecting useful information for the next inference step by modifying the attention mask
• Achieved acceleration in inference stage but observed no significant improvement in reasoning accuracy

Large Language Models Reasoning

Peking University
Sept 2023 – Feb 2024

• Contributed to the Chain of Images (CoI) approach, leveraging generated images to assist models in reasoning through complex textual problems
• Constructed datasets across multiple domains, generated reasoning-supportive images, and fine-tuned models to improve performance on CoI evaluation tasks

Research Outputs

Simulating Human-like Daily Activities with Desire-driven Autonomy
(Under Submission)

• Yiding Wang, Yuxuan Chen, Fangwei Zhong, Long Ma, Yizhou Wang

Enhancing Efficiency in LLM Fine-Tuning via Distributed Parameter Directions
(In Preparation)

Chain of Images for Intuitively Reasoning

• Fanxu Meng, Haotong Yang, Yiding Wang, Muhan Zhang
  arxiv.org/abs/2311.09241

Contest

Neurips 2024 Concordia Contest, Team Member of BIGAI (Rank #9/25)
Oct 2024 – Nov 2024

• Proposed the high-level design of the agent, organizing its functionality into three core modules: perception, decision-making, and persuasion
• Implemented and tested specific modules and scenarios to ensure the agent’s effective performance in cooperative environments

Teaching

Machine Learning (Prof. Muhan Zhang), Teaching Assistant
2024 Fall

Awards

• Peking University Scholarship (¥4000 RMB), 2024
• Yanchuang Capital Scholarship (¥8000 RMB), 2023
• Peking University Excellent Student Award, 2023

Technologies

Languages: C++, C, CUDA, Python