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。
50 篇论文 · 按点赞排序
Noam Issachar, Guy Yariv, Sagie Benaim +3 authors
Dynamic Position Extrapolation (DyPE) enhances ultra-high-resolution image generation by dynamically adjusting positional encodings in pre-trained diffusion transformers, achieving state-of-the-art fidelity without additional sampling cost.
Haochen Wang, Yuhao Wang, Tao Zhang +12 authors
Grasp Any Region (GAR) enhances region-level visual understanding by integrating global contexts and modeling interactions, achieving advanced reasoning and outperforming existing models in captioning and video reference tasks.
Hanxu Hu, Xingxing Zhang, Jannis Vamvas +2 authors
QueST, a framework combining difficulty-aware graph sampling and fine-tuning, generates large-scale synthetic coding problems to enhance the performance of large language models in competitive coding and reasoning tasks.
Chenghao Zhang, Guanting Dong, Xinyu Yang +1 authors
Nyx, a unified mixed-modal retriever, enhances vision-language generation by retrieving and reasoning over mixed-modal data, outperforming existing RAG systems in real-world scenarios.
Heecheol Yun, Kwangmin Ki, Junghyun Lee +1 authors
SAFE, a selective ensembling framework for large language models, improves long-form generation by considering tokenization mismatch and consensus in probability distributions, leading to better accuracy and efficiency.
Liam Parker, Francois Lanusse, Jeff Shen +24 authors
AION-1, a family of large-scale multimodal foundation models, integrates diverse astronomical data using tokenization and transformer-based modeling, achieving strong performance across various downstream tasks.
Byung-Kwan Lee, Ryo Hachiuma, Yong Man Ro +2 authors
A unified reinforcement and imitation learning algorithm creates efficient, lightweight vision-language models that match or exceed leading VLMs in performance.
Yusu Qian, Eli Bocek-Rivele, Liangchen Song +5 authors
Pico-Banana-400K is a large-scale, high-quality dataset for instruction-based image editing, featuring diverse edit pairs, multi-turn editing, preference subsets, and long-short instruction pairs, enabling comprehensive research and benchmarking.
Tian Lan, Bin Zhu, Qianghuai Jia +6 authors
DeepWideSearch is a benchmark that evaluates agents' ability to integrate deep reasoning and wide-scale information collection, revealing significant challenges and limitations in current agent architectures.
Yiqian Yang, Tian Lan, Qianghuai Jia +6 authors
HSCodeComp evaluates deep search agents' hierarchical rule application in predicting product HS Codes, revealing significant performance gaps compared to human experts.
Abdul Fatir Ansari, Oleksandr Shchur, Jaris Küken +20 authors
Chronos-2, a pretrained model with a group attention mechanism, achieves state-of-the-art performance in zero-shot univariate, multivariate, and covariate-informed forecasting tasks.
Shihao Li, Yuanxing Zhang, Jiangtao Wu +20 authors
A new benchmark, IF-VidCap, evaluates video captioning models on instruction-following capabilities, revealing that top-tier open-source models are closing the performance gap with proprietary models.
Shr-Ruei Tsai, Wei-Cheng Chang, Jie-Ying Lee +2 authors
LightsOut enhances Single Image Flare Removal by reconstructing off-frame light sources using a diffusion-based outpainting framework, improving performance across challenging scenarios.
Yuhang Chen, Tianpeng Lv, Siyi Zhang +4 authors
Paper2Web is a benchmark and evaluation framework for academic webpage generation, featuring PWAgent, an autonomous pipeline that enhances content and layout through MCP tools, outperforming end-to-end baselines.
Jiuhai Chen, Le Xue, Zhiyang Xu +12 authors
BLIP3o-NEXT, a unified text-to-image generation and image editing model, uses an Autoregressive + Diffusion architecture to achieve superior performance and realism.
Łukasz Borchmann
The paper advocates for an empiricist approach to evaluating language models, emphasizing frequency of use over traditional theoretical frameworks.
Aleksandr Oganov, Ilya Bykov, Eva Neudachina +7 authors
The Generalized Adversarial Solver improves diffusion model sampling efficiency and quality by combining a simple ODE solver parameterization with adversarial training.
Qianben Chen, Jingyi Cao, Jiayu Zhang +12 authors
A unified framework, A$^2$FM, combines reasoning and agentic capabilities in large language models, improving efficiency and accuracy across benchmarks by adaptively routing queries and optimizing policy.
Mingyu Jo, Jaesik Yoon, Justin Deschenaux +2 authors
Loopholing Discrete Diffusion Models (LDDMs) enhance text generation by preserving distributional information through a deterministic latent pathway, reducing perplexity and improving coherence and performance on reasoning tasks.
Do Xuan Long, Xingchen Wan, Hootan Nakhost +3 authors
VISTA, a multi-agent system, iteratively refines user prompts to enhance video quality and alignment with user intent, outperforming existing methods.
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