Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
50 篇论文 · 按点赞排序
Xinming Wang, Weinong Wang, Hongming Yang +13 authors
Hybrid-thinking multimodal language models suffer from response-pattern misalignment between thinking and non-thinking modes, which is addressed by a diagnostic benchmark and pattern-specific reinforcement learning penalties.
Jian Yang, Haau-Sing Li, Shawn Guo +8 authors
CyberFactory is an open-source framework that builds agentic training data from real vulnerabilities to train Aegis, improving open-weight cybersecurity performance across proof-of-concept generation, patching, and question answering.
Liyan Tang, Cyrus Rashtchian, Chun-Sung Ferng +3 authors
WikiSkill co-evolves reusable agent skills with a persistent knowledge base to systematically accumulate experience and improve performance across models.
Bowen Cui, Weijie Wang, Zeyu Zhang +7 authors
Block3D accelerates text-to-3D generation by using block-wise diffusion with confidence-guided correction to reduce inference time while preserving geometric fidelity.
Zixuan Wang, Yanrui Miao, Zhengxi Lu +6 authors
Agent-G² models hint depth as a Gaussian distribution estimated online from existing rollouts, improving reinforcement learning on long-horizon tasks without extra probing.
Xianyun Sun, Chaoyou Fu, Zhengye Zhang +6 authors
OmniAssistBench evaluates real-time interactive video assistants by reverse-engineering multi-turn interaction videos, revealing that current omni-modal models struggle with visual prompts, context retention, and timely responses.
Zhaochen Yu, Yingcheng Wu, Zhenfei Yin +5 authors
Recuris introduces a recursive memory architecture that tracks progress and guides skill selection to improve long-horizon agent success through localized, validation-gated updates.
Zechen Sun, Zhiwei Zhang, Fei Zhao +7 authors
D³-MOPD dynamically adjusts domain sampling ratios during multi-teacher distillation by monitoring per-domain reverse-KL trajectories, improving convergence efficiency and closing most of the student-to-teacher performance gap.
Bin Dou, Junru Zhang, Zhaoyi Yuan +6 authors
The study proposes a scaling law linking data size to tokenization capacity for user behavior modeling and introduces an adaptive tokenization method to improve efficiency.
Jiaming Zhou, Qihang Zhang, Gangwei Xu +9 authors
Zero-WAM enables robotic manipulation of unseen tasks by conditioning a causal video-action model on in-context human video guidance, supported by an automatically generated dataset and a future-chunk prediction objective.
Hongbo Lu, Liang Yao, Chenghao He +5 authors
RISE adaptively decides when to continue or stop imagination rollouts for planning by weighing expected benefit against cost, supported by a counterfactual driving dataset with expert annotations.
Zhifei Chen, Luozhou Wang, Guibao Shen +8 authors
ReWorld separates short-horizon control and long-horizon memory during training, then bounds both at inference via mixed attention windows, a pose-indexed landmark bank, and distribution-matching LoRA distillation to enable real-time interactive world modeling with strong action fidelity and long-range recall.
Huan-ang Gao, Haohan Chi, Yong Yan +7 authors
Multi-teacher on-policy distillation suffers from token-level budget misallocation across domains, which is addressed by balancing, dynamic allocation, and reward refresh to recover most of the oracle ensemble's capability.
Yunpeng Ba, Zhi Zheng, Yue Xie +7 authors
Evolution strategies improve reasoning diversity and Pass@K over GRPO through sparse functional updates and population diversity, supporting a hybrid training approach.
Yinhao Tang, Youqing Fang, Yanan Sun +8 authors
Mixed supervised fine-tuning on combined reasoning corpora outperforms next-chunk reinforcement learning in efficiency and final accuracy across mathematical and out-of-domain tasks.
Yiming Ren, Xiang Liu, Qumeng Sun +4 authors
AutoResearch is a two-stage autonomous system that grounds research ideas through integrated generation and evidence-based execution to improve experimental reliability and measurable outcomes.
Zhe Liu, Jinghua Hou, Yuxiang Lu +7 authors
StreamPI enhances single-frame vision-language-action models with streaming temporal reasoning via instruction-anchored attention and randomized interval training, improving robot manipulation without extra parameters.
Penghui Qi, Xiangxin Zhou, Wee Sun Lee
BPCO stabilizes critic-based reinforcement learning for language models by combining bounded value predictions, Monte Carlo targets, and adaptive advantage estimation, matching group-based methods with single-response sampling.
Hongbo Liu, Peixian Chen, Sihan Liu +12 authors
A new benchmark evaluates how well multimodal language models follow diverse video-based instructions with visual, audio, and structural constraints.
Wenze Lin, Jiale Zhao, Xitai Jiang +5 authors
OPDVR integrates on-policy distillation with verifiable rewards via a ReLU-gated implicit reward reformulation, improving reasoning performance without extra hyperparameters.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号