arXiv · 2026
GraphThinker: Reinforcing Temporally Grounded Video Reasoning with Event Graph Thinking
Improving temporally grounded video reasoning through reinforcement learning and event-graph-based thinking.
Multimodal learning
Multimodal reasoning
-
-
Pattern Recognition · 2026
Boosting Multi-modal Chain-of-Thought Reasoning by Selective Mixture of Experts
Improving multimodal chain-of-thought reasoning by selectively combining specialised experts.
-
TMLR 2025
TFAR: A Training-Free Framework for Autonomous Reliable Reasoning in Visual Question Answering
Improving visual question answering through training-free reasoning supported by lightweight vision tools.
Vision-language model adaptation
-
ICML 2026
Interpretability Transfer from Language to Vision via Sparse Autoencoders
Using language sparse autoencoders to interpret and intervene on visual concepts in a vision-language model.
-
ICLR 2026
MergeTune: Continued Fine-Tuning of Vision-Language Models
Recovering pretrained knowledge after adaptation through continued fine-tuning of vision-language models.
-
Under review · 2024
Vision-Language Subspace Prompting
Subspace prompting for adapting vision-language models to multimodal tasks.
Selected papers co-authored by Da Li. For the full publication list, visit Google Scholar.