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。
686 篇论文 · 按点赞排序
Namgyu Ho, Huzama Ahmad, Woosung Koh +3 authors
Declarative Attention lets language models declare relevant context regions during reasoning to skip most KV cache reads, reducing attended tokens with small accuracy trade-offs.
Haoran Wei, Yaofeng Sun, Yukun Li
DeepSeek-OCR 2 introduces DeepEncoder V2 that dynamically reorders visual tokens based on semantic content, enabling more human-like causal reasoning in 2D image understanding through cascaded 1D causal structures.
Ilia Semenkov, Daria Kleeva, Ivan Dakhtin +2 authors
Redesigning a MEG-to-audio retrieval model with spherical harmonics, source-space mapping, and input interventions reveals that narrative speech structure, acoustic onsets, and intensity drive retrieval while reducing parameters.
Yalcin Tur, Jalal Naghiyev, Haoquan Fang +4 authors
RD-VLA introduces a recurrent architecture for vision-language-action models that adapts computational depth through latent iterative refinement, achieving constant memory usage and improved task success rates.
Abhigya Verma, Khyati Mahajan, Amit Kumar Saha +4 authors
Vision language models (VLMs) have achieved strong performance on visual document understanding benchmarks such as DocVQA, ChartQA, and MMLongBench-Doc. However, real-world documents combine multiple factors such as length, layout complexity, modality, and question difficulty, which makes it difficult to attribute model failures to specific causes. We introduce SynthDocBench, a fully synthetic benchmark for long-context visual document understanding that systematically controls factors including document length, layout structure, modality composition, and question type. The benchmark is constructed using a combinatorial design, each factor is varied independently across generated documents, enabling controlled analysis of model behavior. Documents are generated end to end using an LLM pipeline across six layout archetypes, with a 40 percent random override to prevent models from exploiting spurious correlations. Additionally, SynthDocBench spans long-context documents with substantially greater length and structural diversity than existing benchmarks. Evaluating seven frontier VLMs, we uncover three failure modes that existing benchmarks cannot surface: sharp degradation with document length, a systematic positional sensitivity in which the middle third of a document is hardest for five of six models and five of six models show a negative Early-to-Late trend (steepest decline: 8.3 percentage points), and breakdown of chart comprehension in long-document settings. These results suggest that current models may be overfitting to benchmark artifacts rather than achieving robust long-context visual document understanding.
Yuyao Zhang, Junjie Gao, Zhengxian Wu +11 authors
Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to track task progress. When search attempts fail to yield useful evidence, current single- and multi-agent systems can become trapped in repetitive loops, wasting search budgets and ultimately compromising the quality and completeness of the final output. We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state. First, we formulate open-domain information seeking as relational schema completion with grounded citations, where agents discover entities, populate attributes across linked tables, and anchor each value to source evidence. Then we design Search-Oriented Context Management (SOCM), which externalizes the evolving state into Frontier Task, an Evidence Graph, a Coverage Map, and Failure Memory. Built on SOCM, SearchOS applies a pipeline-parallel scheduling mechanism that overlaps the execution of sub-agents and continuously refills freed slots with tasks targeting unresolved coverage gaps to improve utilization and throughput. To schedule and control the execution of search agents, SearchOS introduces a Search Tool Middleware Harness that intercepts model and tool interactions to record grounded evidence and react to stalls or budget exhaustion, and provides a reusable hierarchical skill system comprising strategy and access skills to augment the agents' search process and avoid repeating failed search patterns across runs. On WideSearch and GISA, SearchOS leads all metrics among the evaluated single- and multi-agent baselines, paving the way toward robust information-seeking collaboration.
Xiaomi Robotics Team, Jun Guo, Piaopiao Jin +31 authors
We present Xiaomi-Robotics-1, a foundational vision-language-action (VLA) model capable of (1) following diverse language instructions to perform a wide range of mobile manipulation tasks in unseen environments out-of-the-box, and (2) efficiently adapting to novel downstream tasks with minimal fine-tuning data. We propose a two-stage training recipe consisting of pre-training and post-training. During pre-training, we imbue the model with broad and generalizable action-generation capabilities by training on over 100k hours of real-world manipulation trajectories collected via UMI devices. Crucially, we develop a scalable auto-labeling pipeline that annotates trajectory clips with natural languages describing scene state transitions, providing rich and precise conditioning for action learning. During post-training, we aim to align these capabilities with robot embodiments and imperative instructions that humans naturally use to prompt robots. Extensive experiments demonstrate strong scaling behavior. Xiaomi-Robotics-1 consistently improves with increased data scales and model sizes during pre-training. This scaling behavior directly transfers to post-training, where a stronger pre-training model yields better out-of-the-box real-robot performance in unseen environments. Furthermore, Xiaomi-Robotics-1 serves as a strong robot foundation policy that can be efficiently fine-tuned on complex, dexterous tasks with high data efficiency. Across multiple simulation benchmarks, Xiaomi-Robotics-1 outperforms state-of-the-art methods. Notably, it establishes a new state-of-the-art with a 57.6% success rate on RoboCasa365, surpassing the previous best of 46.6%. Furthermore, it achieves an average score of 20.07 on RoboDojo, significantly outperforming the prior state-of-the-art (13.07). Code and model checkpoints will be released. Project page: https://robotics.xiaomi.com/xiaomi-robotics-1.html
Lei Bai, Jiaqi Cao, Chiyu Chen +122 authors
Intern-S2-Preview is a scientific agentic foundation model series that integrates multimodal pre-training, multi-task reinforcement learning, and memory-augmented extensions to support long-horizon scientific reasoning and forecasting.
Zhenxiong Yu, Zhi Yang, Zhiheng Jin +19 authors
Spider-Sense framework provides intrinsic and selective agent security through event-driven defense with intrinsic risk sensing, achieving low attack success and false positive rates with minimal latency overhead.
Guangting Zheng, Yiyuan Zhang, Tao Yang +4 authors
Direct end-to-end training of latent generative models avoids collapse via entropy preservation and asymmetric dynamics, using a generation-first strategy to achieve state-of-the-art image synthesis and unified multimodal generation.
Baorong Shi, Bo Cui, Boyuan Jiang +17 authors
MedXIAOHE is a medical vision-language foundation model that enhances clinical understanding through entity-aware continual pretraining, reinforcement learning, and tool-augmented agentic training for reliable diagnostic reasoning.
Wenkai Yang, Weijie Liu, Ruobing Xie +3 authors
On-policy distillation is extended through a generalized framework that introduces flexible reference models and reward scaling factors, demonstrating improved performance through reward extrapolation and reward correction techniques.
Soohyun Ryu, Sohee Kim, Eunho Yang
Large vision-language models trained on synthetic block-manipulation tasks improve 3D spatial reasoning and generalize to real-world visual tasks.
Zishuo Li, Bowen Yang, Changtao Miao +29 authors
Egocentric videos of human manipulation provide scalable supervision for embodied intelligence, yet existing resources rarely combine low-cost continuous capture, manipulation-level structured annotations, and reusable tools for robot learning. We present Open-AoE, an open, community-oriented egocentric manipulation dataset and toolchain spanning the full pipeline from smartphone capture to model training. Its first release contains approximately 2,000 hours of manipulation video collected in natural environments by 500+ contributors using 400+ smartphones. The dataset provides text annotations, MANO-based hand poses, camera trajectories, and temporally localized atomic actions. Open-AoE further includes a data processing pipeline that transforms raw recordings into structured samples through temporal action segmentation, semantic annotation, hand reconstruction, and camera trajectory reconstruction. Meanwhile, we provide a separate downstream toolchain supports visualization, cross-embodiment retargeting, model-specific data conversion, and training recipes for VLA policies, WAMs, and World Models. By integrating scalable capture, structured processing, and downstream adaptation, Open-AoE reduces the barriers to both data contribution and reuse, providing practical open infrastructure for embodied model training, human-to-robot transfer, and world modeling.
Xiangchen Cheng, Yunwei Jiang, Jianwen Sun +7 authors
A bounded contract approach for long-horizon LLM agents uses typed retrieval to assemble fresh prompts, enabling isolated analysis of memory components and demonstrating improved performance in complex decision-making tasks.
Kang Liao, Yihang Luo, Xiao-Ming Wu +7 authors
Puffin-World is a unified multimodal framework that jointly models physics, geometry, and appearance for physically consistent 3D world generation, reconstruction, and closed-loop exploration.
Geuntaek Lim, Sungjune Park, Jaeyun Lee +5 authors
Video-Oasis diagnostics reveal that half of existing video benchmarks can be solved without visual input, exposing significant capability gaps in current video understanding models.
Sensen Gao, Zhaoqing Wang, Qihang Cao +3 authors
PixWorld presents a unified pixel-space diffusion approach for 3D reconstruction and generation that overcomes limitations of latent-space methods through direct image-level supervision and geometry-aware feature alignment.
Venus Team, Zhuohan Cai, Haoxing Chen +28 authors
UI-Venus-2 is a general-purpose multimodal GUI agent that uses unified reasoning-action loops, expanded environment coverage, and robust verification to enable reliable real-world digital automation.
Yijun Lu, Rui Ye, Jiajun Wang +4 authors
Answer-Backtracked Credit Assignment improves long-horizon search agents by assigning dense step-level rewards via answer-backtracked clue recovery and clue-anchored scoring, yielding strong benchmark results with small models.
Yuran Wang, Siqiao Huang, Mingleyang Li +21 authors
OpenWAM factorizes world-action pretraining into modular components to identify key design principles, yielding a scalable open model with strong simulation and real-robot performance.
Shuailei Ma, Jiaqi Liao, Xinyang Wang +24 authors
LingBot-Video presents a DiT-based video pretraining framework with Mixture-of-Experts architecture, specialized data augmentation, and multi-dimensional reward system for embodied intelligence applications.
Lingao Xiao, Yalun Dai, Yangyu Huang +17 authors
ResearchStudio-Reel automates research dissemination by composing specialized skills around a shared paper extractor, generating consistent and editable artifacts including posters, videos, and blogs with hard pass/fail quality gates.
Zhongwei Yu, Yan Song, Xue Yan +9 authors
A recurrent Large Discovery Model couples generative proposal with a Bayesian non-parametric reward surrogate to guide uncertainty-aware search across molecules, proteins, and programs.
Ziyuan Liu, Hengqi Liu, Zichuan Wang +6 authors
Two large-scale search agents are trained via a multi-stage pipeline combining supervised fine-tuning and reinforcement learning against live search, achieving state-of-the-art open-source results on complex web benchmarks through rigorous trajectory filtering and inference-time context management.
Junlin Han, Shengbang Tong, David Fan +4 authors
Systematic experiments reveal how unified multimodal pretraining transfers knowledge across vision and language, showing that early joint training, shared attention with modality-specific feed-forward layers, and low-compute recipes improve synergy and reduce vision laziness.
Qihao Zhao, Yangyu Huang, Yalun Dai +8 authors
ResearchStudio-Idea provides a skill suite for effective research ideation that combines literature search, novelty checking, and pattern-guided generation to produce traceable research proposals.
Junhao Shen, Zhanqiu Zhang, Yiwen Guo +1 authors
Skill-α uses reinforcement learning with rollback rewards to progressively generate agent skills by evaluating sequential edits against downstream task performance.
Jiale Xu, Rendong Liang, Yuhao Long +4 authors
Meshy T2 is a fast native mesh generation framework using flow matching and a vertex-set mesh VAE to produce high-fidelity, artist-style meshes with interactive speed and controllable complexity.
Jiahao Zhao, Xiaomin Yu, Zhongxiang Sun +5 authors
ToolArtist is an agentic image generation model that unifies reasoning, tool use, and image synthesis under a single policy via supervised fine-tuning and reinforcement learning.
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