Scalable Behaviour Cloning on Browser Using via Skill Distillation

Kaisen Yang$^{*}$, Zheng Jiang$^{*}$, Yuzhao Peng$^{*}$, Houde Qian$^{*}$, Boshi Zhang$^{*}$, Youjie Zheng, Shijin Hong, Qingle Liu, Ruoyu Han, Bohan Lyu, Bingxiang He, Eren Cai, Calvin Xiao, Qinhuai Na$^{\dagger}$

• BrowserBC distills human browser interaction traces into reusable natural-language skills organized as a skill graph, providing retrievable and composable procedural priors for efficient browser-agent execution.

Test-time Scaling over Perception: Resolving the Grounding Paradox in Thinking with Images

Zheng Jiang, Yiming Chen, Nan He, Jiahui Chen, Chaoyang Li, Houde Qian, Lifeng Sun$^{\dagger}$

• TTSP is a test-time scaling framework that resolves the grounding paradox in tool-augmented visual reasoning by scaling perception through parallel exploration, reliability filtering, and iterative knowledge refinement.

SubFLOT: Submodel Extraction for Efficient and Personalized Federated Learning via Optimal Transport

Zheng Jiang, Nan He, Yiming Chen, Lifeng Sun$^{\dagger}$

• SubFLOT is a server-side personalized federated pruning framework that leverages optimal transport and adaptive regularization to address system and statistical heterogeneity without accessing local data.

MedVR: Annotation-Free Medical Visual Reasoning via Agentic Reinforcement Learning

Zheng Jiang$^{*}$, Heng Guo$^{*}$, Chengyu Fang$^{*}$, Changchen Xiao, Xinyang Hu, Lifeng Sun$^{\dagger}$, Minfeng Xu$^{\dagger}$

• MedVR is the first end-to-end reinforcement learning framework that seamlessly integrates visual and textual reasoning for medical VLMs, obviating the need for costly intermediate supervision.

M3Ret: Unleashing Zero-shot Multimodal Medical Image Retrieval via Self-Supervision

Che Liu$^{*}$, Zheng Jiang$^{*}$, Chengyu Fang$^{*}$, Heng Guo, Yanjie Zhou, Jiaqi Qu, Le Lu, Minfeng Xu$^{\dagger}$

• A unified visual encoder without any modality-specific customization for various medical visual modalities in 2D and 3D.

Photon: Speedup Volume Understanding with Efficient Multimodal Large Language Models

Chengyu Fang$^{*}$, Heng Guo$^{*}$, Zheng Jiang, Chunming He, Xiu Li$^{\dagger}$, Minfeng Xu$^{\dagger}$

• Photon is a variable-length 3D medical VQA framework with instruction-conditioned token scheduling and surrogate gradients, achieving adaptive acceleration and state-of-the-art performance.