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。
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.
50 篇论文 · 按点赞排序
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.
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.
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.
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.
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.
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.
Shengbang Tong, David Fan, John Nguyen +18 authors
Controlled multimodal pretraining experiments reveal key insights about unified visual representations, data complementarity, world modeling emergence, and efficient scaling through mixture-of-experts architectures.
Weinan Dai, Hanlin Wu, Qiying Yu +13 authors
CUDA Agent, a large-scale agentic reinforcement learning system, achieves state-of-the-art performance in CUDA kernel optimization by combining scalable data synthesis, skill-augmented development environment, and reinforcement learning techniques.
Yuan Liang, Ruobin Zhong, Haoming Xu +46 authors
SkillNet introduces an open infrastructure for systematically accumulating and transferring AI skills through a unified ontology, significantly improving agent performance across multiple domains.
Ibragim Badertdinov, Maksim Nekrashevich, Anton Shevtsov +1 authors
A large-scale dataset of software engineering tasks spanning multiple programming languages and repositories was created using an automated pipeline that generates executable environments and filters unreliable instances through LLM validation.
Zonglin Yang, Lidong Bing
MOOSE-Star framework enables efficient training and inference for generative reasoning by addressing combinatorial complexity through decomposed subtasks, hierarchical search, and bounded composition.
Zimo Wen, Boxiu Li, Wanbo Zhang +11 authors
Unified multimodal models show mixed performance in generation-to-understanding tasks, with specific subtasks benefiting from enhanced spatial and reasoning capabilities while overall performance lags behind specialized vision-language models.
Ruisheng Cao, Mouxiang Chen, Jiawei Chen +17 authors
Qwen3-Coder-Next is an 80-billion-parameter language model that activates only 3 billion parameters during inference, achieving strong coding capabilities through agentic training with verifiable task synthesis and reinforcement learning.
Qiyuan Zhang, Junyi Zhou, Yufei Wang +8 authors
RubricBench is introduced as a benchmark for evaluating rubric-guided reward models in large language model alignment, addressing the lack of discriminative complexity and ground-truth annotations in existing benchmarks.
Zhenyu Tang, Chaoran Feng, Yufan Deng +5 authors
A new reward model called SpatialScore is introduced to improve spatial relationship understanding in text-to-image generation through reinforcement learning with a large-scale dataset of preference pairs.
Guoxin Chen, Fanzhe Meng, Jiale Zhao +12 authors
Current code agent benchmarks fail to capture real-world complexity, prompting the creation of BeyondSWE to evaluate broader reasoning and knowledge scopes, alongside SearchSWE to study external knowledge integration in coding tasks.
Xinyu Zhu, Yihao Feng, Yanchao Sun +5 authors
A synthetic reasoning dataset called CHIMERA is introduced to overcome data-centric challenges in training large language models for cross-domain reasoning, achieving performance comparable to much larger models.
Maojun Sun, Yue Wu, Yifei Xie +5 authors
A lightweight retrieval model called DARE incorporates data distribution information into function representations to improve R package retrieval, achieving superior performance over existing embedding models while enabling more reliable statistical analysis through an R-oriented LLM agent.
Grigory Arshinov, Aleksandr Boriskin, Sergey Senichev +4 authors
OpenAutoNLU is an open-source automated machine learning library for NLU tasks that employs data-aware training selection and includes integrated diagnostics and LLM features through a minimal low-code interface.
Jiachun Li, Shaoping Huang, Zhuoran Jin +5 authors
MMR-Life benchmark evaluates multimodal large language models' reasoning capabilities across real-life scenarios using diverse multimodal multi-image questions.
Zhaochen Su, Jincheng Gao, Hangyu Guo +10 authors
AgentVista presents a comprehensive benchmark for multimodal agents requiring long-horizon tool interactions across multiple modalities and complex real-world scenarios.
Shengqu Cai, Weili Nie, Chao Liu +8 authors
A training paradigm combining mode seeking and mean seeking in a Decoupled Diffusion Transformer enables efficient generation of high-quality long videos by leveraging both global flow matching and local distribution matching techniques.
Yang Cao, Feize Wu, Dave Zhenyu Chen +3 authors
VGGT-Det enables sensor-geometry-free multi-view indoor 3D object detection by integrating a Visual Geometry Grounded Transformer encoder with attention-guided query generation and query-driven feature aggregation to effectively utilize semantic and geometric priors.
Weicai Yan, Yuhong Dai, Qi Ran +6 authors
Proact-VL is a multimodal framework that enables real-time interactive AI companions for gaming scenarios with low-latency responses and strong video understanding capabilities.
Yinghao Ma, Haiwen Xia, Hewei Gao +9 authors
A comprehensive ecosystem for music reward modeling under compositional multimodal instruction is established through large-scale datasets, unified benchmark, and parameter-efficient reward models that correlate well with human judgments.
Junjie Fang, Wendi Chen, Han Xue +7 authors
RoboPocket enables efficient robot-free policy iteration using smartphone-based augmented reality visualization and asynchronous online fine-tuning to improve data and sample efficiency in imitation learning.
Qiyuan Zhang, Yufei Wang, Tianhe Wu +5 authors
Generative Reward Models can be improved by structuring Chain-of-Thought reasoning into breadth and depth components and optimizing them through supervised fine-tuning and reinforcement learning with verifiable rewards.
Jiejun Tan, Zhicheng Dou, Liancheng Zhang +3 authors
MemSifter is a framework that uses a small proxy model to offload memory retrieval from large language models, employing reinforcement learning with task-performance rewards and training techniques like curriculum learning and model merging to improve long-term memory efficiency and accuracy.
Yichen Liu, Donghao Zhou, Jie Wang +9 authors
HiFi-Inpaint generates high-fidelity human-product images using shared enhancement attention and detail-aware loss with a new 40K-image dataset.
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