CV

CHEN Yanxin

Ph.D. Student in Artificial Intelligence

yanxinchen26@m.fudan.edu.cn
Shanghai, , China

Summary

Ph.D. student in Artificial Intelligence at Fudan University, focusing on long video understanding, video large models, retrieval-augmented generation, and large-scale VLM training infrastructure.

Education

  • Ph.D. in Artificial Intelligence
    Present
    Fudan University

Work Experience

  • Core Contributor
    2026 -
    MOSS-VL
    Large-scale VLM pretraining and SFT workflow.
    • Responsible for four-stage pretraining and the full SFT workflow, including ablation experiments and formal training runs.
    • Experienced with thousand-GPU cluster training and debugging training-framework, data, NCCL communication, and deadlock issues.

Skills

AI Algorithms

  • Large Language Models
  • Video Large Models
  • Long Video Understanding
  • RAG
  • SFT/RL Post-training
  • Linear Attention

Training Infrastructure

  • Megatron-LM
  • TransformerEngine
  • DeepSpeed
  • FlashAttention
  • FlashAttention-3
  • TP/PP/DP Parallelism
  • Activation Recomputation

Engineering

  • Python
  • C++
  • PyTorch
  • Linux/Bash
  • NCCL Debugging
  • Cluster Debugging

Publications

  • ProEchoMem: Enhancing Long Video Understanding via Multi-Trace Probe-Echo Memory
    2026
    SIGIR 2026
    Co-first author. Proposed the PN-RAG framework for probe generation, structured annotation, and fine-grained reranking in long-video understanding.

Interests

  • Research
    Efficient VLM Training, Long-context Video Modeling, VLM Post-training, Low-precision Training