Yuchong Wu

Yuchong Wu

Undergraduate Student at USTC

Curriculum Vitae

YuChong Wu

Contact

  • Phone: +86 15819499387
  • Email: wuyuchong0317@gmail.com
  • Address: Room 1307, Residence Hall 6, USTC High-tech Campus

Education

University of Science and Technology of China (USTC) B.S. in Computer Science and Technology, Hua Xia Talent Program (Honors) Sep 2022–Present

Weighted Average Score: 90.25/100 — Rank: 17/215

Honors & Awards

  • JAC NIO Scholarship (2023–2024) Awarded to top 5% of students for outstanding academic performance
  • Merit Student Scholarship (2022–2023) Granted for consistent academic excellence and active participation in extracurricular activities

Research & Project Experience

Undergraduate Graduation Thesis, USTC

LLM-based Agentic workflow for Time series anomaly detection Sep 2025–Dec 2025

  • Developed a human-inspired tool-augmented LLM framework for anomaly detection in time series data, integrating external tools for enhanced reasoning and analysis.
  • Achieved state-of-the-art performance, significantly outperforming strong baselines across major metrics.

Undergraduate Research Program, USTC

Exploring Reinforcement Learning Methods for Better Planning and Alignment in LLM Training Dec 2024–Sep 2025

  • Proposed a novel DPO variant leveraging final-layer embedding similarity to selectively identify informative tokens for optimization, mitigating likelihood displacement.

Project Experience

Kunpeng Ascend Special Training Camp, Huawei

July 2024

  • Participated in intensive training on deploying AI models for edge computing applications.
  • Designed and implemented a real-time object detection system using YOLOv5 on the MindSpore framework, deployed on Ascend GPU and Orange Pi.

Robogame Competition, USTC

Mar 2024–Sep 2024

  • Led development of circuit control system and vision system (focused on traditional CV algorithms) for autonomous robotic cars.

Technical Skills

  • Deep Learning: PyTorch, HuggingFace Transformers/Trl, Verl; supervised fine-tuning, RLHF and RLVR on LLMs
  • Agent Training: Agent-lightning framework; familiar with agentic reinforcement learning methodologies
  • Reinforcement Learning: Q-learning, policy gradient methods, actor-critic algorithms with focus on LLM reasoning and alignment
  • Systems: CPU/GPU architectures, CUDA programming (CUDA C++), parallel computation