Summary

Focused on post-training algorithms, especially reinforcement learning, and on optimizer design and implementation. Main tools include PyTorch and TensorFlow for reproduction and engineering work.

Education

• Computer Science and Technology (B.S.)

  2027-07
  Beijing University of Posts and Telecommunications

  GPA: 3.76/4.00

  Courses: Ranked 20/360 in major, CET-6: 503, CET-4: 571

Skills

Portfolio

• SPPO: Sequence-Level PPO for Long-Horizon Reasoning

  2026
  Research

  Addresses temporal credit assignment and memory overhead in long-horizon reasoning by reformulating reasoning as a sequence-level contextual bandit and introducing a decoupled scalar value function.

  View Project
• Anchored Policy Optimization (APO)

  2026
  Research

  Proposes a support-constrained rectification method for exploration collapse and entropy collapse in RLVR, emphasizing support coverage and stable training.

  View Project
• PyTorch Implementation: Spectral-Sphere-Optimizer (SSO)

  2026
  Engineering

  Implemented and merged a PyTorch version of the SSO optimizer for direct use in training frameworks and reproducible experiments.

  View Project