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。
606 篇论文 · 按点赞排序
Shengming Yin, Zekai Zhang, Zecheng Tang +11 authors
A diffusion model for decomposing RGB images into semantically disentangled RGBA layers enables consistent image editing through independent manipulation of individual layers.
Tianyu Fu, Yi Ge, Yichen You +6 authors
Roads to Rome (R2R) selectively utilizes large language models for critical reasoning tasks to enhance efficiency and performance in lightweight models.
Yu Gao, Lixue Gong, Qiushan Guo +28 authors
Seedream 3.0 improves Chinese-English bilingual image generation by enhancing data training, pre-training techniques, and post-training aesthetics, resulting in higher visual quality and faster image generation.
Shivalika Singh, Yiyang Nan, Alex Wang +10 authors
Measuring progress is fundamental to the advancement of any scientific field. As benchmarks play an increasingly central role, they also grow more susceptible to distortion. Chatbot Arena has emerged as the go-to leaderboard for ranking the most capable AI systems. Yet, in this work we identify systematic issues that have resulted in a distorted playing field. We find that undisclosed private testing practices benefit a handful of providers who are able to test multiple variants before public release and retract scores if desired. We establish that the ability of these providers to choose the best score leads to biased Arena scores due to selective disclosure of performance results. At an extreme, we identify 27 private LLM variants tested by Meta in the lead-up to the Llama-4 release. We also establish that proprietary closed models are sampled at higher rates (number of battles) and have fewer models removed from the arena than open-weight and open-source alternatives. Both these policies lead to large data access asymmetries over time. Providers like Google and OpenAI have received an estimated 19.2% and 20.4% of all data on the arena, respectively. In contrast, a combined 83 open-weight models have only received an estimated 29.7% of the total data. We show that access to Chatbot Arena data yields substantial benefits; even limited additional data can result in relative performance gains of up to 112% on the arena distribution, based on our conservative estimates. Together, these dynamics result in overfitting to Arena-specific dynamics rather than general model quality. The Arena builds on the substantial efforts of both the organizers and an open community that maintains this valuable evaluation platform. We offer actionable recommendations to reform the Chatbot Arena's evaluation framework and promote fairer, more transparent benchmarking for the field
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.
Tianhong Li, Kaiming He
Today's denoising diffusion models do not "denoise" in the classical sense, i.e., they do not directly predict clean images. Rather, the neural networks predict noise or a noised quantity. In this paper, we suggest that predicting clean data and predicting noised quantities are fundamentally different. According to the manifold assumption, natural data should lie on a low-dimensional manifold, whereas noised quantities do not. With this assumption, we advocate for models that directly predict clean data, which allows apparently under-capacity networks to operate effectively in very high-dimensional spaces. We show that simple, large-patch Transformers on pixels can be strong generative models: using no tokenizer, no pre-training, and no extra loss. Our approach is conceptually nothing more than "Just image Transformers", or JiT, as we call it. We report competitive results using JiT with large patch sizes of 16 and 32 on ImageNet at resolutions of 256 and 512, where predicting high-dimensional noised quantities can fail catastrophically. With our networks mapping back to the basics of the manifold, our research goes back to basics and pursues a self-contained paradigm for Transformer-based diffusion on raw natural data.
Xiangxin Zhou, Zichen Liu, Haonan Wang +5 authors
A variational reasoning framework treats thinking traces as latent variables, optimizing them through variational inference to improve language model reasoning.
David Dinkevich, Matan Levy, Omri Avrahami +2 authors
Story2Board generates expressive storyboards from natural language using a consistency framework that enhances coherence and diversity without fine-tuning.
Junteng Liu, Yuanxiang Fan, Zhuo Jiang +12 authors
SynLogic, a data synthesis framework, enhances the logical reasoning capabilities of Large Language Models through RL, achieving state-of-the-art performance and improving generalization across various domains.
Junhao Cheng, Yuying Ge, Yixiao Ge +2 authors
AnimeGamer uses Multimodal Large Language Models (MLLMs) and a video diffusion model to generate contextually consistent, dynamic, and high-quality video clips for an infinite game simulation.
Gallil Maimon, Avishai Elmakies, Yossi Adi
A method for training high-quality Speech Language Models on a single GPU within 24 hours, demonstrating cost-effectiveness and outperforming predicted computational scaling laws.
Diankun Wu, Fangfu Liu, Yi-Hsin Hung +1 authors
Spatial-MLLM improves spatial reasoning in multimodal large language models using a dual-encoder architecture with pretrained 2D and 3D structure encoders, achieving state-of-the-art performance on visual spatial tasks.
OpenAI, Ahmed El-Kishky, Alexander Wei +22 authors
General-purpose reinforcement learning applied to large language models outperforms domain-specific systems in complex coding and reasoning tasks, achieving top results in competitions without hand-crafted strategies.
Mistral-AI, Abhinav Rastogi, Albert Q. Jiang +97 authors
Magistral, a scalable reinforcement learning pipeline, demonstrates that RL can enhance multimodal understanding and instruction following in large language models without requiring existing RL traces.
Bingquan Dai, Li Ray Luo, Qihong Tang +9 authors
MeshCoder reconstructs complex 3D objects from point clouds into editable Blender Python scripts, enhancing shape-to-code reconstruction and 3D shape understanding through a multimodal large language model.
Guo Chen, Zhiqi Li, Shihao Wang +18 authors
Eagle 2.5 improves long-context multimodal learning through Automatic Degrade Sampling and Image Area Preservation techniques, enhancing VLMs for video and image understanding, and matches state-of-the-art performance on benchmarks.
Rujun Han, Yanfei Chen, Zoey CuiZhu +15 authors
TTD-DR, a diffusion-based framework, generates high-quality research reports by iteratively refining a preliminary draft with external information and self-evolutionary algorithms, outperforming existing deep research agents.
Fanxu Meng, Zengwei Yao, Muhan Zhang
TransMLA converts GQA-based models to MLA to reduce KV cache size and improve efficiency without sacrificing expressiveness.
Hongcheng Gao, Zihao Huang, Lin Xu +11 authors
The Turing Eye Test evaluates MLLMs' perceptual capabilities on synthetic images, revealing that vision tower generalization is a key gap compared to human perception.
Gheorghe Comanici, Eric Bieber, Mike Schaekermann +3300 authors
Gemini 2.X model family, including Gemini 2.5 Pro and Flash, offers superior coding, reasoning, and multimodal understanding capabilities across a range of computational efficiencies.
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.
Jiaqi Tang, Jianmin Chen, Wei Wei +7 authors
A novel framework, Robust-R1, enhances multimodal large language models' robustness to visual degradations through explicit modeling, supervised fine-tuning, reward-driven alignment, and dynamic reasoning depth scaling, achieving state-of-the-art performance on real-world degradation benchmarks.
Weichen Fan, Haiwen Diao, Quan Wang +2 authors
Unified Autoencoding combines semantic and pixel-level information through a frequency-band modulator, resulting in a latent space with state-of-the-art performance on image benchmarks.
Haoqi Yuan, Yu Bai, Yuhui Fu +6 authors
Being-0, a hierarchical framework integrating Foundation Models with a modular skill library and a lightweight vision-language model, achieves efficient real-time performance on complex humanoid robot tasks by coordinating high-level cognition with low-level control.
Xiangyu Zhao, Peiyuan Zhang, Kexian Tang +7 authors
RISEBench evaluates Reasoning-Informed Visual Editing by assessing temporal, causal, spatial, and logical reasoning in multimodal models like GPT-4o-Native.
Yuxuan Luo, Zhengkun Rong, Lizhen Wang +3 authors
A diffusion transformer framework, DreamActor-M1, enhances image-based human animation through hybrid guidance, achieving robust control, multi-scale adaptability, and long-term temporal coherence.
Alexander Groshev, Anastasiia Iashchenko, Pavel Paramonov +2 authors
GHOST 2.0, consisting of an Aligner and Blender module, achieves state-of-the-art results in head swapping by preserving identity information, handling extreme poses, and seamlessly integrating the reenacted head into the target background.
Prateek Chhikara, Dev Khant, Saket Aryan +2 authors
Mem0, a memory-centric architecture with graph-based memory, enhances long-term conversational coherence in LLMs by efficiently extracting, consolidating, and retrieving information, outperforming existing memory systems in terms of accuracy and computational efficiency.
Junying Chen, Zhenyang Cai, Pengcheng Chen +5 authors
ShareGPT-4o-Image and Janus-4o enable open research in photorealistic, instruction-aligned image generation through a large dataset and multimodal model.
Omkar Thawakar, Dinura Dissanayake, Ketan More +12 authors
A framework for evaluating and improving step-by-step visual reasoning in large language models using a specialized benchmark and a novel multimodal model trained with curriculum learning.
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