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。
Björn Engdahl, Adrian Kosowski, Jan Chorowski +6 authors
A 150M-parameter reasoning model using recurrent latent reasoning and in-context learning achieves a new cost-accuracy frontier on ARC-AGI-1.
682 篇论文 · 按点赞排序
DeepReinforce Team, Xiaoya Li, Xiaofei Sun +4 authors
GrandCode is a multi-agent reinforcement learning system that outperforms human competitors in competitive programming challenges by orchestrating specialized agent modules and employing novel reward policy optimization techniques.
Jianlyu Chen, Yuyang Hu, Hongjin Qian +8 authors
DisCo is a research agent that distills operational knowledge into reusable skills, significantly improving autonomous ML research performance across benchmarks.
Maijunxian Wang, Ruisi Wang, Juyi Lin +53 authors
A large-scale video reasoning dataset and benchmark are introduced to study video intelligence capabilities beyond visual quality, enabling systematic analysis of spatiotemporal reasoning and generalization across diverse tasks.
Kimi Team, Tongtong Bai, Yifan Bai +399 authors
Kimi K3 is a large-scale mixture-of-experts model with native vision and long-context capabilities that improves scaling efficiency and achieves strong performance across coding, reasoning, and agentic tasks.
Hongyuan Adam Lu, Z. L., Victor Wei +5 authors
A novel framework for improving large language model performance through textual frequency analysis, including laws, distillation, and curriculum training approaches.
Yihao Wang, Yuheng Ji, Mingyu Cao +54 authors
Orca establishes a unified world latent space through next-state-prediction modeling using multimodal data and demonstrates superior performance in downstream tasks compared to specialized baselines.
Ming Qian, Tianjian Ouyang, Mingchao Sun +25 authors
ABot-Earth 0.5 generates realistic 3D environments from satellite imagery using 3D Gaussian Splatting representation, enabling fast synthesis and real-time visualization for Embodied AI applications.
Ke Yang, Chenglong Wang, Michel Galley +4 authors
StudentSim trains personalized student simulators from sparse data to mirror learner responses and adapt to tutor guidance, outperforming existing models across chess, writing, and math.
Hongyuan Adam Lu, Z. L. Victor Wei, Qun Zhang +28 authors
Looped World Models introduce iterative latent state refinement through shared transformer blocks, achieving 100x parameter efficiency while adapting computational depth to prediction complexity.
Ziheng Qin, Yaxin Lu, Zhangyang Atlas Wang +1 authors
StateM is a runtime system that improves long-horizon agent execution through durable states, recoverable runbooks, and enforceable procedural controls without altering model weights.
Fangfu Liu, Kai He, Tianchang Shen +7 authors
A generative multi-agent world model is presented that uses simplex rotary agent encoding and sparse hub attention to enable scalable, permutation-symmetric interaction between multiple agents in interactive video generation.
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.
Xiyuan Yang, Sheikh Sarwar, Jingru Cheng +8 authors
World Model RL replaces costly environment execution with a learned world model and applies debiasing and denoising to accelerate post-training of autonomous research agents.
NeoHorse Team, Guoliang Cao, Guohao Dai +34 authors
NeoHorse-1 uses agentic post-training with intelligent routing, structured feedback loops, and curriculum-based distillation to improve model capabilities across agent benchmarks.
Yiyou Sun, Xinyang Han, Weichen Zhang +305 authors
Agents' Last Exam (ALE) is a benchmark for evaluating AI agents on long-term, economically valuable real-world tasks across 13 industry clusters with 1K+ tasks, revealing significant gaps between benchmark performance and practical deployment.
Yuntian Deng, Pengyu Nie, Stuart Shieber
Compile by training converts natural-language specifications into reusable neural functions by distilling teacher-generated examples into small adapters, enabling efficient deployment without remote model dependencies.
Xin Zhou, Zongchuang Zhao, Zhibo Yang +13 authors
Qwen-Drive-1.0 is a vision-language foundation model for autonomous driving that unifies 3D perception, visual question answering, and motion planning via shared representations and staged training.
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.
Hao Liang, Zhengyang Zhao, Meiyi Qiang +22 authors
DataFlex is a unified framework for dynamic data-centric training of large language models that supports sample selection, domain mixture adjustment, and sample reweighting while maintaining compatibility with standard training workflows and enabling efficient large-scale deployment.
Haoquan Fang, Jiafei Duan, Donovan Clay +26 authors
MolmoAct2 presents an open-action reasoning model for robotics that improves upon previous systems through specialized vision-language-model backbones, new datasets, open-weight action tokenizers, architectural redesign for continuous-action prediction, and adaptive reasoning for reduced latency.
Shaobo Wang, Xuan Ouyang, Tianyi Xu +9 authors
OPUS is a dynamic data selection framework that improves pre-training efficiency by scoring data candidates based on optimizer-induced update projections in a stable proxy-derived target space, achieving superior performance with reduced computational overhead.
Chiyu Ma, Shuo Yang, Kexin Huang +7 authors
FIPO enhances reinforcement learning for language models by using discounted future-KL divergence to improve credit assignment and extend reasoning chains, achieving better mathematical problem-solving performance.
Tianle Zeng, Hanxuan Chen, Yanci Wen +1 authors
CARLA-Air integrates high-fidelity driving and multirotor flight simulation within a unified Unreal Engine framework, supporting joint air-ground agent modeling with photorealistic environments and multi-modal sensing capabilities.
Mind Lab, Vin Bo, Asher Cai +73 authors
Macaron-V1 is an open agent-model family that uses a Mixture-of-LoRA architecture and recursive self-improvement to enable continual learning and collaboration across specialized tasks.
Weiliang Chen, Haowen Sun, Jun Gao +40 authors
HarnessEval-W uses hierarchical sub-agents to decompose world-model evaluations into verifiable reasoning chains that justify scores with transparent evidence.
Qihan Ren, Peng Wang, Ruikun Cai +8 authors
Supervised finetuning and reinforcement learning exhibit conditional cross-domain generalization in reasoning tasks, influenced by optimization dynamics, data quality, and model capability, with asymmetric outcomes between reasoning improvement and safety degradation.
I. Apanasevich, M. Artemyev, R. Babakyan +25 authors
Green-VLA is a five-stage vision-language-action framework for real-world robot deployment that achieves generalization across different robot embodiments through multimodal training and reinforcement learning.
Guan Wang, Changling Liu, Chenyu Wang +6 authors
A Hierarchical Recurrent Model architecture with specialized training on instruction-response pairs achieves competitive language modeling performance with significantly reduced computational requirements compared to traditional Transformer-based approaches.
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.
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