🔬 Research Experience

Meta AI - Research Scientist Intern May 2025 – Oct. 2025, California

Research topic: Adaptive Reasoning for Video-Language Models

  • Developed the first auto-thinking video models (8B), reducing inference latency by 70% while improving reasoning capability.
  • Proposed a “think once, answer twice” training paradigm, coupled with confidence-based early-exit inference.
  • Built scalable RL-based training and evaluation pipelines for video benchmarks, achieving SOTA performance.
  • VideoAuto-R1 was accepted to CVPR 2026.