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。
Jingqi Tong, Mingzhe Li, Hangcheng Li +20 authors
Great scientists have strong judgement and foresight, closely tied to what we call scientific taste. Here, we use the term to refer to the capacity to judge and propose research ideas with high potential impact. However, most relative research focuses on improving an AI scientist's executive capability, while enhancing an AI's scientific taste remains underexplored. In this work, we propose Reinforcement Learning from Community Feedback (RLCF), a training paradigm that uses large-scale community signals as supervision, and formulate scientific taste learning as a preference modeling and alignment problem. For preference modeling, we train Scientific Judge on 700K field- and time-matched pairs of high- vs. low-citation papers to judge ideas. For preference alignment, using Scientific Judge as a reward model, we train a policy model, Scientific Thinker, to propose research ideas with high potential impact. Experiments show Scientific Judge outperforms SOTA LLMs (e.g., GPT-5.2, Gemini 3 Pro) and generalizes to future-year test, unseen fields, and peer-review preference. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than baselines. Our findings show that AI can learn scientific taste, marking a key step toward reaching human-level AI scientists.
50 篇论文 · 按点赞排序
Ruisi Wang, Zhongang Cai, Fanyi Pu +11 authors
Diffusion-based video models demonstrate reasoning capabilities through denoising steps rather than frame sequences, exhibiting behaviors like working memory, self-correction, and perception-before-action within specialized transformer layers.
Jian Yang, Wei Zhang, Jiajun Wu +25 authors
InCoder-32B is a 32-billion-parameter code model trained on industrial datasets with extended context length and execution verification to improve performance in hardware-aware programming tasks.
Tianyu Xie, Jinfa Huang, Yuexiao Ma +9 authors
SocialOmni presents a benchmark for evaluating social interactivity in omni-modal large language models across speaker identification, interruption timing, and natural interruption generation, revealing gaps between perceptual accuracy and conversational competence.
Lei Huang, Xiang Cheng, Chenxiao Zhao +6 authors
Language feedback is leveraged in reinforcement learning to improve exploration efficiency and sample utilization through grouped critique aggregation and joint generation-refinement optimization.
Zhixia Zhang, Zixuan Huang, Xin Xia +7 authors
HACRL enables collaborative reinforcement learning where heterogeneous agents share verified rollouts during training to improve collectively while maintaining independent operation at inference time, with HACPO achieving superior performance through efficient sample utilization and cross-agent knowledge transfer.
Kimi Team, Guangyu Chen, Yu Zhang +34 authors
Residual connections with PreNorm are standard in modern LLMs, yet they accumulate all layer outputs with fixed unit weights. This uniform aggregation causes uncontrolled hidden-state growth with depth, progressively diluting each layer's contribution. We propose Attention Residuals (AttnRes), which replaces this fixed accumulation with softmax attention over preceding layer outputs, allowing each layer to selectively aggregate earlier representations with learned, input-dependent weights. To address the memory and communication overhead of attending over all preceding layer outputs for large-scale model training, we introduce Block AttnRes, which partitions layers into blocks and attends over block-level representations, reducing the memory footprint while preserving most of the gains of full AttnRes. Combined with cache-based pipeline communication and a two-phase computation strategy, Block AttnRes becomes a practical drop-in replacement for standard residual connections with minimal overhead. Scaling law experiments confirm that the improvement is consistent across model sizes, and ablations validate the benefit of content-dependent depth-wise selection. We further integrate AttnRes into the Kimi Linear architecture (48B total / 3B activated parameters) and pre-train on 1.4T tokens, where AttnRes mitigates PreNorm dilution, yielding more uniform output magnitudes and gradient distribution across depth, and improves downstream performance across all evaluated tasks.
Shenghai Yuan, Yuanyang Yin, Zongjian Li +3 authors
Helios is a 14 billion parameter autoregressive diffusion model for video generation that achieves real-time performance and high-quality long-video synthesis without conventional optimization techniques.
MiroMind Team, S. Bai, L. Bing +41 authors
MiroThinker-1.7 and MiroThinker-H1 are research agents that enhance complex reasoning through structured planning, contextual reasoning, and tool interaction, with MiroThinker-H1 incorporating verification at local and global levels for more reliable multi-step problem solving.
Yujia Zhang, Xiaoyang Wu, Yunhan Yang +6 authors
Utonia enables cross-domain point cloud representation learning through a unified self-supervised transformer encoder, enhancing perception and supporting embodied and multimodal reasoning tasks.
Kaijin Chen, Dingkang Liang, Xin Zhou +4 authors
Hybrid Memory enables video world models to maintain consistent tracking of dynamic subjects during occlusion by combining archival storage for static backgrounds with active tracking for moving objects, using a specialized architecture with tokenized memory and spatiotemporal retrieval mechanisms.
Yinjie Wang, Xuyang Chen, Xiaolong Jin +2 authors
OpenClaw-RL framework enables policy learning from diverse next-state signals across multiple interaction modalities using asynchronous training with PRM judges and hindsight-guided distillation.
Yawen Luo, Xiaoyu Shi, Junhao Zhuang +5 authors
ShotStream enables real-time interactive multi-shot video generation through causal architecture design, dual-cache memory mechanisms, and two-stage distillation to maintain visual consistency and reduce latency.
Daxiang Dong, Mingming Zheng, Dong Xu +17 authors
Qianfan-OCR is a 4B-parameter vision-language model that unifies document parsing, layout analysis, and understanding while maintaining strong performance across multiple OCR benchmarks through its Layout-as-Thought mechanism.
Junyoung Seo, Hyunwook Choi, Minkyung Kwon +10 authors
What if a world simulation model could render not an imagined environment but a city that actually exists? Prior generative world models synthesize visually plausible yet artificial environments by imagining all content. We present Seoul World Model (SWM), a city-scale world model grounded in the real city of Seoul. SWM anchors autoregressive video generation through retrieval-augmented conditioning on nearby street-view images. However, this design introduces several challenges, including temporal misalignment between retrieved references and the dynamic target scene, limited trajectory diversity and data sparsity from vehicle-mounted captures at sparse intervals. We address these challenges through cross-temporal pairing, a large-scale synthetic dataset enabling diverse camera trajectories, and a view interpolation pipeline that synthesizes coherent training videos from sparse street-view images. We further introduce a Virtual Lookahead Sink to stabilize long-horizon generation by continuously re-grounding each chunk to a retrieved image at a future location. We evaluate SWM against recent video world models across three cities: Seoul, Busan, and Ann Arbor. SWM outperforms existing methods in generating spatially faithful, temporally consistent, long-horizon videos grounded in actual urban environments over trajectories reaching hundreds of meters, while supporting diverse camera movements and text-prompted scenario variations.
Zhanhui Zhou, Lingjie Chen, Hanghang Tong +1 authors
A unified open-source framework is presented that standardizes core components of diffusion language modeling for reproduction, customization, and accessible development of both large and small models.
Yukang Cao, Haozhe Xie, Fangzhou Hong +4 authors
HSImul3R presents a unified framework for 3D reconstruction of human-scene interactions that bridges the perception-simulation gap through physics-grounded bidirectional optimization and reinforcement learning.
Yulin Li, Tengyao Tu, Li Ding +5 authors
ReBalance is a training-free framework that balances reasoning in large models by using confidence indicators to detect and correct overthinking and underthinking behaviors through dynamic steering vectors.
Yiying Yang, Wei Cheng, Sijin Chen +5 authors
OmniLottie framework generates high-quality vector animations from multi-modal instructions using a specialized Lottie tokenizer and pretrained vision-language models.
Yuwen Du, Rui Ye, Shuo Tang +4 authors
Deep search capabilities have become an indispensable competency for frontier Large Language Model (LLM) agents, yet the development of high-performance search agents remains dominated by industrial giants due to a lack of transparent, high-quality training data. This persistent data scarcity has fundamentally hindered the progress of the broader research community in developing and innovating within this domain. To bridge this gap, we introduce OpenSeeker, the first fully open-source search agent (i.e., model and data) that achieves frontier-level performance through two core technical innovations: (1) Fact-grounded scalable controllable QA synthesis, which reverse-engineers the web graph via topological expansion and entity obfuscation to generate complex, multi-hop reasoning tasks with controllable coverage and complexity. (2) Denoised trajectory synthesis, which employs a retrospective summarization mechanism to denoise the trajectory, therefore promoting the teacher LLMs to generate high-quality actions. Experimental results demonstrate that OpenSeeker, trained (a single training run) on only 11.7k synthesized samples, achieves state-of-the-art performance across multiple benchmarks including BrowseComp, BrowseComp-ZH, xbench-DeepSearch, and WideSearch. Notably, trained with simple SFT, OpenSeeker significantly outperforms the second-best fully open-source agent DeepDive (e.g., 29.5% v.s. 15.3% on BrowseComp), and even surpasses industrial competitors such as Tongyi DeepResearch (trained via extensive continual pre-training, SFT, and RL) on BrowseComp-ZH (48.4% v.s. 46.7%). We fully open-source the complete training dataset and the model weights to democratize frontier search agent research and foster a more transparent, collaborative ecosystem.
Shiva Krishna Reddy Malay, Shravan Nayak, Jishnu Sethumadhavan Nair +6 authors
Large language models are shifting from passive information providers to active agents intended for complex workflows. However, their deployment as reliable AI workers in enterprise is stalled by benchmarks that fail to capture the intricacies of professional environments, specifically, the need for long-horizon planning amidst persistent state changes and strict access protocols. In this work, we introduce EnterpriseOps-Gym, a benchmark designed to evaluate agentic planning in realistic enterprise settings. Specifically, EnterpriseOps-Gym features a containerized sandbox with 164 database tables and 512 functional tools to mimic real-world search friction. Within this environment, agents are evaluated on 1,150 expert-curated tasks across eight mission-critical verticals (including Customer Service, HR, and IT). Our evaluation of 14 frontier models reveals critical limitations in state-of-the-art models: the top-performing Claude Opus 4.5 achieves only 37.4% success. Further analysis shows that providing oracle human plans improves performance by 14-35 percentage points, pinpointing strategic reasoning as the primary bottleneck. Additionally, agents frequently fail to refuse infeasible tasks (best model achieves 53.9%), leading to unintended and potentially harmful side effects. Our findings underscore that current agents are not yet ready for autonomous enterprise deployment. More broadly, EnterpriseOps-Gym provides a concrete testbed to advance the robustness of agentic planning in professional workflows.
Mohamad Zbib, Mohamad Bazzi, Ammar Mohanna +2 authors
Speculative decoding effectiveness depends on draft model training data alignment with downstream tasks, with specialized drafters performing better when combined through confidence-based routing rather than simple averaging.
Jiyuan Wang, Chunyu Lin, Lei Sun +8 authors
RL3DEdit uses reinforcement learning with rewards from a 3D foundation model to achieve multi-view consistent 3D editing from 2D editing priors.
Peng Xia, Jianwen Chen, Xinyu Yang +10 authors
A continual meta-learning framework for large language model agents that jointly evolves policies and reusable behavioral skills while minimizing downtime through opportunistic updates and skill-driven adaptation.
Hejun Dong, Junbo Niu, Bin Wang +3 authors
MinerU-Diffusion is a diffusion-based framework that replaces autoregressive decoding with parallel diffusion denoising for document OCR, improving robustness and decoding speed.
Xiangyan Qu, Zhenlong Yuan, Jing Tang +9 authors
Image-CoT methods are extended to image editing with ADE-CoT, which improves efficiency and performance through adaptive resource allocation, edit-specific verification, and opportunistic stopping mechanisms.
Yicheng Zou, Dongsheng Zhu, Lin Zhu +171 authors
Intern-S1-Pro is a one-trillion-parameter scientific multimodal foundation model that enhances general and scientific capabilities through advanced agent functionalities and specialized task mastery across multiple scientific disciplines.
SII-GAIR, Sand. ai, Ethan Chern +42 authors
daVinci-MagiHuman is an open-source audio-video generative model that synchronizes text, video, and audio through a single-stream Transformer architecture, achieving high-quality human-centric content generation with efficient inference capabilities.
Qinsi Wang, Hancheng Ye, Jinhee Kim +12 authors
Structure of Thought prompting technique enhances language model performance by guiding explicit intermediate text structuring across diverse tasks, while T2S-Bench benchmark evaluates and improves text-to-structure capabilities with comprehensive scientific domain coverage.
Boqiang Zhang, Lei Ke, Ruihan Yang +5 authors
Penguin-VL demonstrates that text-only initialized vision encoders can achieve superior performance in multimodal understanding tasks compared to traditional contrastive pretraining methods, enabling efficient deployment on resource-constrained devices.
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