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
Yujin Kim, Namgyu Ho, Sangmin Hwang +7 authors
LLM-as-a-Tutor framework extends LLM role from judge to tutor by dynamically adjusting prompt difficulty through pairwise comparison and constraint addition, improving instruction-following performance in reinforcement learning.
Yifei Shen, Bo Li, Xinjie Zhang
A minimal viable pipeline for skill optimization is proposed through Zeroth-Order optimization formalization, eliminating redundancies while maintaining convergence and generalization through trajectory exploration, consensus mining, and validation gating principles.
Zhekai Chen, Chengqi Duan, Kaiyue Sun +4 authors
UniClawBench introduces a capability-driven benchmark for evaluating proactive agents in real-world environments using live Docker container evaluation and closed-loop assessment with multiple agent roles.
Zhenyu Hou, Yujiang Li, Jie Tang +1 authors
Asynchronous reinforcement learning with single-rollout optimization addresses stability issues in LLM training for complex tasks, outperforming existing methods in coding and reasoning benchmarks.
Jisen Li, Bingxuan Li, Nanyi Jiang +10 authors
paper pilot is a multi-turn literature search agent that uses executable workflows to improve search accuracy and reduce errors by incorporating user feedback and controlled workflow corruption training.
Xinyan Chen, Ziyu Guo, Renrui Zhang +2 authors
OpenCoF framework introduces a reasoning video dataset and model that improve temporal reasoning through diverse supervision and explicit reasoning tokens for visual and textual cues.
Yi Pan, Miao Pan, Qi Lu +8 authors
VLA-Corrector addresses limitations of action chunking in vision-language-action models by introducing a lightweight latent-space vision monitor that enables adaptive corrective replanning, improving robustness in contact-rich manipulation tasks.
Anthony Hu, Václav Volhejn, Adrien Ramanana Rahary +24 authors
A large-scale multiplayer world model trained on extensive gameplay data demonstrates stable long-horizon rollouts in a complex physics-based environment while maintaining coherence across multiple agents' actions.
Wenzheng Zeng, Siyi Jiao, Chen Gao +2 authors
A parallelized autoregressive framework for dense video captioning that improves generation efficiency by exploiting weak local dependencies across temporally distinct events while maintaining temporal grounding accuracy.
Haoxiang Ma, Junhao Cai, Xiaoxu Xu +26 authors
InternVLA-A1.5 integrates pretrained vision-language models with future prediction in latent space to enable efficient robot manipulation with preserved semantics and long-horizon execution.
Heqing Yang, Yang Yi, Liyao Wang +8 authors
EVA-Client is an open-source framework that unifies real-robot policy deployment, data collection, and evaluation through a component-decoupled architecture with inspectable execution workflows.
Haozhan Tang, Zerui Wang, Yuxian Gu +2 authors
A novel zero-shot method called Jet-Long enables efficient long-context processing for large language models by dynamically adapting rescaling factors and utilizing a bifocal attention mechanism that maintains high performance across varying sequence lengths.
Zhifeng Kong, Sang-gil Lee, Jaehyeon Kim +17 authors
A unified audio-text large language model is presented that integrates audio and text processing through a shared transformer decoder, achieving superior performance across multiple audio and speech tasks while maintaining strong text reasoning capabilities.
Chang Nie, Jiaju Wei, Junlan Feng +2 authors
Light-Omni is a multimodal agent framework that enables efficient video understanding through dual contextual states, achieving faster and more accurate video processing by eliminating iterative reasoning while maintaining semantic alignment.
Lukas Hauzenberger, Niklas Schmidinger, Anamaria-Roberta Hartl +5 authors
KVpop learns optimal key-value cache eviction by directly supervising keep-or-drop decisions using future-attention targets, achieving high performance with reduced memory usage.
Suhyeong Park, Junha Jung, Jungwoo Park +1 authors
Object-aware token merging framework SaMer compresses image-side tokens while preserving query-selectable visual evidence, achieving significant storage reduction and improved retrieval performance.
Gawon Seo, Dongwon Kim, Suha Kwak
ACID is a decision-time planning framework that enhances action-conditioned world models by enforcing cycle action consistency to improve trajectory realism and reduce computational requirements.
Haoran Feng, Ruiyang Zhang, Longyi Zhang +2 authors
Canvas360 is a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with fine-tuning, featuring a large-scale dataset and novel modeling techniques for improved geometric consistency and global coherence.
Ali Motahharynia, Mohammadreza Ghaffarzadeh-Esfahani, Mahsa Sheikholeslami +4 authors
DrugGen-2 generates small molecules conditioned on disease ontology and target protein sequences through fine-tuning GPT-2 with supervised learning and reinforcement learning using GRPO, achieving superior molecular diversity and binding affinity compared to baseline models.
Haofei Xu, Rundi Wu, Philipp Henzler +7 authors
A minimalist pixel-space diffusion transformer using plain ViT architecture directly processes 3D point map patches conditioned on image tokens from DINOv3, outperforming complex latent-based models while maintaining simplicity and robustness in ambiguous regions.
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