CV

Ming Ma

Ph.D. Student @ CAS

mam2022@ion.ac.cn
+86 13854225898
Shanghai, CN

Summary

Ph.D. student at the Center for Excellence in Brain Science and Intelligence Technology, CAS. Research focused on Computational Neuroscience and LLM Interpretability.

Education

  • Computational Neuroscience
    Present
    Chinese Academy of Sciences (CAS)
  • Automation
    2022.06
    Shandong University
  • Automation
    2020.06
    Beijing Institute of Technology
  • Mechanical and Electronic Engineering
    2019.03
    Naval Aviation University

Work Experience

  • Research Intern (Omni Base Team)
    2026.01 - Present
    Alibaba Tongyi Lab
    Multi-modal Agent framework and Long-video QA optimization.
    • Built a multi-agent system based on Planner-Critic-Reflect for long-video understanding.
    • Developed a multi-modal long-term memory system based on Mem0 and LightMem.
  • Research Intern (Ling Base Team)
    2025.10 - 2026.01
    Ant Group
    Pre-training data quality and structured data strategies.
    • Optimized knowledge conflict resolution in pre-training corpora.
    • Proposed adaptive conversion strategies for structured table data.
  • Research Intern
    2025.07 - 2025.10
    Microsoft Research Asia (MSRA)
    LLM Multi-Agent Systems and Knowledge Exploration.
    • Proposed DoVer, an intervention-driven auto-debugging framework for LLM multi-agent systems.
    • Developed Co-evolving Dual-Graph for open-ended deep research.

Publications

  • DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems (ICLR 2026)
    2026
  • QAQ: Bidirectional Semantic Coherence for Selecting High-Quality Synthetic Code Instructions
    2026
  • A Tale of Two Graphs: Separating Knowledge Exploration from Outline Structure for Open-Ended Deep Research
    2026
  • Label Words as Local Task Vectors in In-Context Learning
    2024
  • SimLens for Early Exit in Large Language Models: Eliciting Accurate Latent Predictions with One More Token
    2025
  • From end-to-end to step-by-step: Learning to abstract via abductive reinforcement learning (IJCAI 2025)
    2025