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。
DeepSeek-AI, Daya Guo, Dejian Yang +197 authors
DeepSeek-R1-Zero and DeepSeek-R1 utilize reinforcement learning and multi-stage training to enhance reasoning capabilities, with DeepSeek-R1 achieving performance comparable to OpenAI-o1-1217.
50 篇论文 · 按点赞排序
Kimi Team, Angang Du, Bofei Gao +91 authors
A multi-modal LLM trained with reinforcement learning achieves state-of-the-art reasoning performance across various benchmarks by utilizing long context scaling and effective policy optimization methods.
Kuang-Huei Lee, Ian Fischer, Yueh-Hua Wu +4 authors
Mind Evolution, an evolutionary search strategy using a language model, outperforms other inference methods in natural language planning tasks by generating, recombining, and refining candidate responses.
Siyu Yuan, Zehui Chen, Zhiheng Xi +3 authors
Agent-R, an iterative self-training framework using Monte Carlo Tree Search, enhances large language models' ability to recover from errors and correct actions in real-time within interactive environments.
Boqiang Zhang, Kehan Li, Zesen Cheng +12 authors
A vision-centric multimodal foundation model, VideoLLaMA3, improves image and video understanding through strategic training stages and adaptive vision token encoding.
Yilun Zhao, Lujing Xie, Haowei Zhang +16 authors
A comprehensive benchmark for evaluating foundation models in video understanding includes domain-specific questions and expert evaluations, highlighting the gap between current models and human expertise.
Zhenran Xu, Longyue Wang, Jifang Wang +7 authors
FilmAgent, a multi-agent collaborative framework based on LLMs, automates film production processes including scriptwriting, cinematography, and actor positioning, outperforming other models in human evaluations.
Alsu Sagirova, Yuri Kuratov, Mikhail Burtsev
Shared Recurrent Memory Transformer (SRMT) enhances cooperation in multi-agent reinforcement learning through implicit information exchange, outperforming baselines in navigation tasks and generalizing to unseen scenarios.
Zihan Qiu, Zeyu Huang, Bo Zheng +7 authors
Calculating Load-Balancing Loss using a global-batch instead of a micro-batch improves performance and domain specialization in MoE-based large language models.
Yujia Qin, Yining Ye, Junjie Fang +32 authors
UI-TARS, a native GUI agent model using screenshots as input, outperforms commercial models in various benchmarks through enhanced perception, unified action modeling, system-2 reasoning, and iterative training with reflective online traces.
Yafu Li, Xuyang Hu, Xiaoye Qu +2 authors
TPO refines LLM outputs in real-time using textual critiques, achieving preference alignment without retraining.
Yichen He, Guanhua Huang, Peiyuan Feng +4 authors
A quality Paper Search agent, PaSa, uses reinforcement learning on synthetic and real academic queries, outperforming Google and GPT-4-based systems.
Zibo Zhao, Zeqiang Lai, Qingxiang Lin +68 authors
Hunyuan3D 2.0, an advanced 3D synthesis system, uses flow-based diffusion transformers for shape generation and diffusion priors for texture synthesis, outperforming existing models in geometry, condition alignment, and texture quality.
Zhenghao Lin, Zihao Tang, Xiao Liu +31 authors
Sigma, a large language model for the system domain, uses DiffQKV attention to enhance inference efficiency and achieves exceptional performance compared to GPT-4 on a new benchmark called AIMicius.
Daniel Garibi, Shahar Yadin, Roni Paiss +6 authors
TokenVerse uses a DiT-based text-to-image model to disentangle and personalize complex visual concepts from multiple images using an optimization-based framework.
Yuhang Zang, Xiaoyi Dong, Pan Zhang +10 authors
IXC-2.5-Reward is a multi-modal reward model that enhances Large Vision Language Models by aligning them with human preferences, improving performance across various tasks and applications.
Ang Lv, Ruobing Xie, Yining Qian +5 authors
Autonomy-of-Experts (AoE) improves MoE models by enabling experts to autonomously select inputs based on self-evaluated capacity, reducing the need for routers and enhancing performance.
Jiwen Yu, Yiran Qin, Xintao Wang +3 authors
GameFactory leverages pre-trained video diffusion models and a multi-phase training strategy to generate diverse, action-controllable game videos with scene generalization.
Ziyu Guo, Renrui Zhang, Chengzhuo Tong +4 authors
Chain-of-Thought reasoning techniques enhance autoregressive image generation by improving verification, aligning model preferences, and using specialized reward models like PARM and PARM++.
Jie Liu, Gongye Liu, Jiajun Liang +15 authors
VideoReward and alignment algorithms improve video generation by addressing unsmooth motion and misalignment using human feedback in a unified reinforcement learning framework.
Maciej Besta, Julia Barth, Eric Schreiber +15 authors
A comprehensive blueprint for modularizing reasoning language models (RLMs) to enhance accessibility and scalability, incorporating diverse reasoning structures, RL concepts, and supervision schemes.
Tairan Fu, Javier Conde, Gonzalo Martínez +2 authors
LLMs are more confident in their answers when reasoning is provided before the answer, regardless of correctness, due to modifications in probability estimation.
Haotian Luo, Li Shen, Haiying He +6 authors
O1-Pruner, an RL-style fine-tuning method, reduces inference time and enhances accuracy in long-thought reasoning LLMs by optimizing token usage.
Zhenhailong Wang, Haiyang Xu, Junyang Wang +5 authors
A hierarchical multi-agent framework called Mobile-Agent-E enhances mobile task navigation through self-evolution, improving performance and efficiency on complex long-horizon tasks.
Zhongwei Ren, Yunchao Wei, Xun Guo +4 authors
VideoWorld, an auto-regressive video generation model trained from visual data, acquires complex knowledge including rules, reasoning, and planning, and achieves high performance in tasks like Video-Go and robotic control without traditional reinforcement learning techniques.
Hongjin Su, Ruoxi Sun, Jinsung Yoon +3 authors
Learn-by-interact, a data-centric framework, synthesizes interaction data for LLM-based agents in various environments, improving performance in tasks like coding, web browsing, and desktop activity through backward construction and optimized retrieval methods.
Chao He, Jianqiang Ren, Liefeng Bo
Textoon is a method that generates diverse 2D cartoon characters in the Live2D format using textual descriptions and advanced language and vision models.
Kairui Hu, Penghao Wu, Fanyi Pu +5 authors
Video-MMMMU evaluates Large Multimodal Models' knowledge acquisition from videos through perception, comprehension, and adaptation stages, revealing performance gaps compared to humans.
Rui Li, Xiaohan Wang, Yuhui Zhang +2 authors
Temporal Preference Optimization (TPO) enhances video-LMMs' temporal grounding through preference learning, improving performance on long-form video benchmarks with fewer manual annotations.
Sili Chen, Hengkai Guo, Shengnan Zhu +4 authors
Video Depth Anything provides high-quality, consistent depth estimation for long videos using an efficient spatial-temporal head and a temporal consistency loss, achieving state-of-the-art performance in zero-shot video depth estimation.
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