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。
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.
50 篇论文 · 按点赞排序
Hanzhang Zhou, Panrong Tong, Xu Zhang +13 authors
Qwen-UI-Agent is a foundation GUI agent that integrates mobile, computer, web, and search environments with CLI execution, long-horizon online reinforcement learning, and autonomous data generation to achieve state-of-the-art real-world device performance.
Bing Yan, Gregory Wolfe, Stefano Martiniani +1 authors
Chemistry literature synthesis often requires assembling specific findings scattered across many publications, yet existing literature-search systems primarily return ranked document lists. As a result, scientists and AI agents need to locate relevant information, verify their provenance, and assemble cross-paper answers manually. We present AskChem, a claim-centered infrastructure for cross-paper chemistry search. AskChem changes the unit of retrieval from the paper to the provenance-carrying claim: each paper is converted into atomic, typed claims, each grounded by a source DOI and a verbatim quote or an explicit evidence locator. Over this shared claim store, AskChem exposes complementary structures for search and synthesis: a stabilized faceted taxonomy for hierarchical retrieval and browsing, an evidence graph linking claims through relations, and an exploratory living taxonomy that situates indexed papers under scientific principles. AskChem currently indexes 2.4M claims from 147K papers and provides a web interface, as well as REST, SDK, and MCP access for AI agents. On AskChem-Bench, grounding a GPT-5.5 reader in AskChem yields 100% resolvable DOIs, compared with 88.3% without retrieval, and the highest citation density among five tested systems. AskChem is live at https://askchem.org.
Zeyu Zhang, Ziliang Guo, Yihang Sun +14 authors
Metis introduces memory foundation models that embed persistent, dynamically evolving native memory states and autonomous storage procedures directly into foundation models via memory attention and gradient-free updates.
Jianshu Zhang, Keliang Wu, Haoran Lu +8 authors
A survey unifies progress reward modeling for robotic learning by analyzing model interfaces, internal construction methods, and evaluation benchmarks.
Junlin Yang, Che Jiang, Yu Fu +21 authors
OpenMLE enables recursive self-improvement in machine learning engineering via a full-stack system combining execution-grounded operator learning and long-horizon evolutionary search, significantly boosting benchmark performance with a 35B model.
Shuyao Shang, Yuqi Wang, Ruopeng Gao +4 authors
PhiZero learns a discrete physical-language representation from videos to explicitly reason about world dynamics before rendering future frames, improving coherence and enabling interactive simulation.
Simple AI, Yuteng Wei, Jinming Ma +15 authors
High-fidelity robot-free UMI data enables direct real-robot deployment without teleoperation anchors, improving manipulation policy accuracy and success rates.
Hengyi Xie, Chenfei Yao, Xianjin Wu +7 authors
TurboVLA replaces the LLM-centric vision-language-action pipeline with direct vision-language interaction and a compact decoder to cut inference costs while maintaining high robotic manipulation performance.
Yunlong Lin, Zixu Lin, Zhaohu Xing +23 authors
JarvisHub is a canvas-based agent framework that supports sustained, inspectable multimodal creative production through editable project states and progressive agent planning.
Zhongming Yu, Hengjia Yu, Boqin Yuan +12 authors
CodeNib provides reusable multi-view repository indexing with faster updates and reduced context tokens while maintaining accurate navigation across code edits.
Jiangnan Li, Yuqing Li, Mo Yu +2 authors
RARG integrates relevance scores into direct corpus search to prioritize documents and excerpts, accelerating accurate evidence retrieval for complex questions.
Jiaxing Li, Kai Zou, Cindy Zhou +7 authors
Joint distillation aligns initialization and refinement distributions to improve coverage and diversity in autoregressive video generation.
Bo-Wen Zhang, Junwei He, Wen Wang +5 authors
CoRT improves rubric-based reinforcement learning by using counterfactual replay to assign token-level credit weights within responses, boosting performance without auxiliary scorers.
Junlin Liu, Jiangwang Chen, Zixin Song +7 authors
Multi-Agent Protocol Distillation improves agentic search by using structured, style-normalized protocols to transfer reasoning from proprietary teachers to smaller models via dense distillation combined with sparse reinforcement learning.
Bingnan Li, Haozhe Wang, Haozhong Xiong +5 authors
On-policy diffusion distillation under classifier-free guidance suffers from branch-level error compensation, which is resolved by separately matching positive predictions and guidance directions.
Siyao Li, Jiawei Gu, Shuai Liu +15 authors
HumanCLAW decouples high-level vision-language decisions from low-level motor execution to evaluate embodied action intelligence, revealing that current vision-language models lack embodied self-awareness.
Haodong Li, Tianfei Ren, Xiaoxiao Ma +25 authors
VideoCoCo uses executable Blender code as an intermediate reasoning step to generate physically consistent videos via simulation and draft-conditioned editing.
Jiangwang Chen, Zixin Song, Junlin Liu +10 authors
DecoEvo improves open-ended LLM optimization by separately evolving a solver and a rubric generator to expose hidden weaknesses without relying on fixed evaluation criteria.
Jooyeol Yun, Jintae Park, Hyesu Lim +3 authors
ReDesign is an agentic framework that reconstructs editable design files from images by composing specialized tools with stepwise verification, achieving high editability across layout, color, and text modifications.
Yan Yang, Xiangru Jian, Ziyang Luo +7 authors
StateAct improves computer-use agents by grounding actions, verification, and memory in direct program state rather than screenshots, boosting success rates while reducing cost.
Rubin Wei, Jiaqi Cao, Jiarui Wang +4 authors
Scaling a separate parametric memory module to billions of parameters improves language model performance more efficiently than scaling the base model alone.
Hao Liang, Qifeng Cai, Yibo Lin +11 authors
DataPrep-Bench unifies evaluation of LLM-driven data construction and quality assessment via downstream training utility across multiple domains, introducing a skill-guided construction agent and a distribution-based quality metric.
Qixun Wang, Yang Shi, Letian Cheng +11 authors
Beacon improves agentic visual reasoning by enhancing adaptive tool use and ensuring tools genuinely extend capabilities on hard tasks without harming easy ones.
Pengyu Wang, Benfeng Xu, Shaohan Wang +4 authors
Controlled scaling experiments across nested corpus tiers show lexical retrieval overtakes agentic search at scale, favoring global ranking over sequential exploration.
Lai Wei, Chengqi Li, Jiapeng Li +3 authors
CLBench-V is a multimodal benchmark evaluating context learning across grounding, application, and knowledge acquisition, revealing substantial gaps in current models.
Siyuan Huang, Pengyu Cheng, Haotian Liu +10 authors
Skill Self-Play co-evolves a task proposer, solver, and dynamic skill controller via reinforcement learning to balance verifiable execution with open-ended task diversity, improving LLM performance on reasoning and tool-use benchmarks.
Yukang Cao, Haozhe Xie, Beichen Wen +13 authors
ACE is a human-centric recording system that captures synchronized multisensory home interaction data to support embodied AI and imitation learning.
Yuran Wang, Zekun Wang, Bohan Zeng +10 authors
Flux-OPD stabilizes evolving contextual supervision for open-ended language model distillation by decomposing reverse KL divergence into geometric mean targets and conflict terms.
Yu Wang, Yi-Kai Zhang, Wentao Shi +8 authors
CAST improves long-horizon game training by deriving turn-level credit from solver value changes and injecting them into reinforcement learning with verifiable rewards.
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