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。
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.
50 篇论文 · 按点赞排序
Kimi Team, Tongtong Bai, Yifan Bai +323 authors
Kimi K2.5 is an open-source multimodal agentic model that enhances text and vision processing through joint optimization techniques and introduces Agent Swarm for parallel task execution.
Haifeng Wang, Hua Wu, Tian Wu +435 authors
ERNIE 5.0 is a production-scale trillion-parameter autoregressive model that unifies multimodal understanding and generation through sparse MoE architecture and elastic training.
Dawei Zhu, Rui Meng, Yale Song +4 authors
_paperbanana is an agentic framework that automates the creation of publication-ready academic illustrations using advanced vision-language models and image generation techniques.
Wenxuan Huang, Yu Zeng, Qiuchen Wang +12 authors
Vision-DeepResearch introduces a multimodal deep-research paradigm enabling multi-turn, multi-entity, and multi-scale visual and textual search with deep-research capabilities integrated through cold-start supervision and reinforcement learning.
Yifei Wang, Yueqi Wang, Zhenrui Yue +6 authors
FASA is a novel framework that uses query-aware token eviction and functional sparsity in RoPE to reduce KV cache memory usage while maintaining high performance in long-context LLM tasks.
Yu Zeng, Wenxuan Huang, Zhen Fang +13 authors
Vision-DeepResearch benchmark addresses limitations in evaluating visual-textual search capabilities of multimodal models by introducing realistic evaluation conditions and improving visual retrieval through multi-round cropped-search workflow.
Ximing Lu, David Acuna, Jaehun Jung +12 authors
Golden Goose synthesizes unlimited RLVR tasks from unverifiable internet text by creating multiple-choice question-answering versions of fill-in-the-middle tasks, enabling large-scale training and achieving state-of-the-art results in cybersecurity and other domains.
Zelai Xu, Zhexuan Xu, Ruize Zhang +7 authors
Multi-agent systems using reinforcement learning enable parallel information seeking with scalable orchestration, achieving performance comparable to larger single agents.
Jian Chen, Yesheng Liang, Zhijian Liu
DFlash is a speculative decoding framework that uses a lightweight block diffusion model for parallel token drafting, achieving significant speedup over existing autoregressive methods while maintaining high-quality outputs.
Yuling Shi, Chaoxiang Xie, Zhensu Sun +7 authors
Multimodal large language models can effectively understand source code when represented as compressed images, achieving significant token reduction while maintaining or improving performance on code comprehension tasks.
Jianhao Ruan, Zhihao Xu, Yiran Peng +8 authors
AOrchestra is a framework-agnostic agentic system that uses a tuple-based abstraction to dynamically create specialized task executors, achieving improved performance on complex benchmarks through automated agent creation and resource management.
Johannes Kirmayr, Lukas Stappen, Elisabeth André
Current LLM agent benchmarks fail to evaluate reliability in real-world scenarios with uncertain user inputs, prompting the creation of CAR-bench to test consistency, uncertainty management, and capability awareness in in-car assistant applications.
Jane Luo, Chengyu Yin, Xin Zhang +10 authors
RPG-Encoder framework transforms repository comprehension and generation into a unified cycle by encoding code into high-fidelity Repository Planning Graph representations that improve understanding and reconstruction accuracy.
Jinyuan Li, Chengsong Huang, Langlin Huang +4 authors
Training multimodal process reward models efficiently through balanced-information scoring that prioritizes label mixture and reliability while achieving full-data performance with only 10% of training data.
Dianyi Wang, Chaofan Ma, Feng Han +8 authors
UniReason integrates text-to-image generation and image editing through a dual reasoning paradigm that enhances planning with world knowledge and uses editing for visual refinement, achieving superior performance on reasoning-intensive benchmarks.
Liyan Xu, Mo Yu, Fandong Meng +1 authors
Research investigates latent planning dynamics in large language models through a probing method called Tele-Lens, revealing limited global planning and enabling improved uncertainty estimation and CoT bypass recognition.
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.
Jiefeng Chen, Bhavana Dalvi Mishra, Jaehyun Nam +3 authors
MARS is a modular AI research automation framework that uses budget-aware planning, modular construction, and reflective memory to achieve state-of-the-art performance in autonomous machine learning research.
Zhixue Fang, Xu He, Songlin Tang +5 authors
3DiMo enables view-agnostic human motion control in video generation by training a motion encoder alongside a pretrained video generator to distill driving frames into compact motion tokens that align with the generator's spatial priors.
Haozhen Zhang, Quanyu Long, Jianzhu Bao +4 authors
MemSkill introduces a learnable and evolvable memory system for LLM agents that dynamically selects and refines memory operations through controller-executor-designer components.
Andrei Panferov, Erik Schultheis, Soroush Tabesh +1 authors
Quantized training method Quartet II improves NVFP4 format utilization for large language model pre-training through enhanced gradient estimation and faster GPU execution.
Mouxiang Chen, Lei Zhang, Yunlong Feng +15 authors
A scalable framework for constructing real-world software engineering environments from GitHub pull requests using an efficient building agent with self-verification and hacking detection capabilities.
Xiaoyu Tian, Haotian Wang, Shuaiting Chen +12 authors
ASTRA is an automated framework that trains tool-augmented language models using synthetic data and verifiable reinforcement learning to improve multi-step decision-making capabilities.
Fanfan Liu, Youyang Yin, Peng Shi +3 authors
Research analyzes RLVR algorithms' impact on response length in LLMs and VLMs, proposing LUSPO to eliminate length bias and improve reasoning performance.
Mohan Jiang, Dayuan Fu, Junhao Shi +8 authors
Large language models face challenges in long-horizon agentic workflows due to lack of authentic long-dependency training data, which is addressed by leveraging pull request sequences for structured supervision through progressive decomposition, consistency enforcement, and refinement from bug-fix histories.
Chiwei Zhu, Benfeng Xu, Mingxuan Du +4 authors
A file-system-based dual-agent framework enables large language model agents to perform extended research tasks beyond context window limitations by using persistent storage as external memory.
Yue Ding, Yiyan Ji, Jungang Li +12 authors
OmniSIFT is a modality-asymmetric token compression framework for Omni-LLMs that reduces computational overhead through spatio-temporal video pruning and vision-guided audio selection while maintaining superior performance.
Bohan Zeng, Kaixin Zhu, Daili Hua +24 authors
Current world models lack unified frameworks despite task-specific advances, necessitating a comprehensive approach integrating interaction, perception, symbolic reasoning, and spatial representation.
Yizhao Gao, Jianyu Wei, Qihao Zhang +11 authors
Hybrid Sparse Attention architecture interleaves full and sparse attention layers, using full attention output to guide sparse layer token selection and cache reuse for improved efficiency and performance.
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