On-policy distillation (OPD) supervises a student on its own trajectories with token-level signals from a frozen teacher, yet how a sampled loss allocates updates across tokens remains poorly understood. We analyze the gradient of the per-token K2 estimator of reverse KL with respect to the student logits. The $\ell_1$...
Bing Shao, Jia-Zheng Zhang, Long Ma et al.· 5 citations
ReDraft (Reference-Driven Revision and Fine-Tuning) shows that revising the model's own rollout rather than directly imitating an expert trajectory can reconcile cold-start acquisition with prior-capability retention.
Zhi-Hao Zhang, Ming-Qi Wu, Qiaole Dong et al.· 0 citations
Experiments on document parsing benchmarks show that SCVER improves robustness under reduced input resolution and achieves a better accuracy-efficiency trade-off, demonstrating the effectiveness of on-demand visual evidence retrieval for fine-grained perception.
Ming-Xu Chai, Chen-Yu Liu, Zi-Yu Shen et al.· 0 citations
PACE (Planning with Adaptive Cognitive Effort), a framework that enables interleaved reasoning and execution through two key innovations: an Interleaved Think-Act architecture that pipelines cognitive processing with action execution, and a Dynamic Budget Allocator that adapts reasoning token budgets to available execu...
Yuchen Huang, Xijiang Ying, Zhenhua Ma et al.· 1 citation
Sci-MMR is introduced, a benchmark for multi-step evidence-grounded scientific reasoning built on structured argument graphs linking scientific claims, citation-grounded knowledge, visual evidence, and supporting regions, and it is found that current answer-centric benchmarks substantially overestimate the evidence-gro...
Jia-Qiang Li, Ya-Jie Yang, Zhi-Heng Xi et al.· 0 citations
This work adapt structured expert judgment from decision theory, using context-aware calibration questions to estimate expert reliability based on the quality of its probabilistic predictions, and employs Cooke-style log weighting, which penalises overconfident incorrect predictions and favours well-calibrated experts.
The Prefix-Adaptive Block Diffusion Model (PA-BDM) is proposed, which replaces intra-block bidirectional denoising with causal denoising from prefix to suffix and treats the block size as a maximum candidate range rather than a fixed commitment unit.
Ming-Xu Chai, Zi-Yu Shen, Chen-Yu Liu et al.· arXiv.org· 0 citations
CAFE (Coupled Agent--Feedback Evolution), a framework in which a shared-parameter model alternates between search-agent and critic roles, is introduced, suggesting that a self-improving search agent needs feedback that co-evolves with the policy it guides.
Bo-Yang Liu, Sen-Jie Jin, Pei-Xin Wang et al.· 1 citation
AgentGym2 is presented, a new evaluation framework with task instances grounded in real-world end-to-end working demands that measures agents'ability to execute end-to-end procedures, discover tools via exploration, compose tools for unseen tasks, and remain robust to noisy and underspecified information.
Zhiheng Xi, Dingwen Yang, Jiaqi Liu et al.· Annual Meeting of the Associ...· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
