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。
Omer Bar-Tal, Hila Chefer, Omer Tov +11 authors
A text-to-video diffusion model using Space-Time U-Net architecture generates realistic, diverse, and coherent videos through a single pass, achieving state-of-the-art results and supporting various content creation and editing tasks.
50 篇论文 · 按点赞排序
Fanghua Yu, Jinjin Gu, Zheyuan Li +6 authors
SUPIR, an advanced image restoration method combining generative prior and model scaling, achieves high-quality results and flexible restoration guided by textual prompts.
Daya Guo, Qihao Zhu, Dejian Yang +10 authors
DeepSeek-Coder, a series of open-source code models trained from scratch, outperforms both open-source and closed-source models across multiple benchmarks.
Mehmet Saygin Seyfioglu, Karim Bouyarmane, Suren Kumar +2 authors
Diffuse to Choose is a diffusion-based inpainting model that balances real-time performance with high-fidelity detail preservation and semantic accuracy.
Lihe Yang, Bingyi Kang, Zilong Huang +3 authors
Depth Anything is a robust monocular depth estimation model built on a large-scale dataset with data augmentation and auxiliary supervision strategies, achieving state-of-the-art results on various datasets and enhancing depth-conditioned ControlNet.
Tianle Cai, Yuhong Li, Zhengyang Geng +4 authors
Medusa enhances Large Language Model inference by adding parallel decoding heads and using tree-based attention to predict multiple tokens simultaneously, achieving significant speedup with minimal latency.
Junxiong Wang, Tushaar Gangavarapu, Jing Nathan Yan +1 authors
MambaByte, a token-free byte-level autoregressive model, demonstrates computational efficiency and competitive performance compared to subword token-based models, with the added benefit of fast inference due to linear length scaling.
Duzhen Zhang, Yahan Yu, Chenxing Li +4 authors
A survey examining the design, performance, and future directions of MultiModal Large Language Models (MM-LLMs) to enhance reasoning and MM task capabilities.
Abhimanyu Hans, Avi Schwarzschild, Valeriia Cherepanova +5 authors
A novel LLM detector using contrasting scores from two pre-trained models achieves high accuracy in identifying machine-generated text without any training data.
Keming Lu, Bowen Yu, Chang Zhou +1 authors
Ditto, a self-alignment method, enhances the role-playing capabilities of open-source large language models by fine-tuning them on a large generated dataset, achieving performance comparable to proprietary chatbots.
Mirac Suzgun, Adam Tauman Kalai
Meta-prompting enhances language model performance by organizing and integrating multiple expert instances to handle complex tasks, demonstrating superiority over traditional methods in zero-shot settings.
Hongliang He, Wenlin Yao, Kaixin Ma +5 authors
WebVoyager, a Large Multimodal Model, enhances web agent performance by handling multiple modalities and real-world web interactions, achieving high success rates in practical tasks.
Haoran Wei, Lingyu Kong, Jinyue Chen +6 authors
Vary-toy, a smaller version of Vary using Qwen-1.8B, enhances vision vocabulary through object detection and achieves competitive performance on visual question answering tasks.
Boyuan Chen, Zhuo Xu, Sean Kirmani +6 authors
A system for training Vision Language Models with internet-scale 3D spatial reasoning data improves their spatial understanding and enables novel applications in quantitative spatial reasoning and robotics.
Ling Yang, Zhaochen Yu, Chenlin Meng +3 authors
A text-to-image generation/editing framework, RPG, uses multimodal LLMs for chain-of-thought reasoning to enhance the compositionality and performance of diffusion models, particularly in handling complex text prompts.
Yinan Zhang, Eric Tzeng, Yilun Du +1 authors
An algorithm using Reinforcement Learning improves text-to-image diffusion models by aligning with human preferences, enhancing sample quality, compositionality, and diversity.
Ge Zhang, Xinrun Du, Bei Chen +20 authors
CMMMU evaluates large multimodal models in college-level Chinese tasks across six disciplines, highlighting gaps and promoting expert-level artificial intelligence development.
Letian Fu, Long Lian, Renhao Wang +6 authors
By using only cross-attention for masked patch reconstruction, CrossMAE achieves comparable and sometimes superior performance to MAE with significantly reduced computational cost.
Penghui Qi, Xinyi Wan, Guangxing Huang +1 authors
A novel scheduling strategy achieves zero pipeline bubbles in synchronous training by splitting backward computation and bypassing synchronizations during optimization, significantly improving throughput.
Elron Bandel, Yotam Perlitz, Elad Venezian +9 authors
Unitxt is a modular and customizable library for preparing and evaluating textual data for generative language models, promoting collaboration and flexibility in the field.
Katherine Crowson, Stefan Andreas Baumann, Alex Birch +3 authors
The Hourglass Diffusion Transformer achieves state-of-the-art performance on high-resolution image generation by combining the scalability of Transformers with the efficiency of U-Nets, operating directly in pixel space without additional high-resolution training techniques.
Zhihong Chen, Maya Varma, Jean-Benoit Delbrouck +14 authors
A large-scale instruction-tuning dataset and an instruction-tuned foundation model with a clinical large language model and vision encoder are introduced to automate CXR interpretation, outperforming existing models on clinical tasks and evaluated for fairness.
Zachary Novack, Julian McAuley, Taylor Berg-Kirkpatrick +1 authors
DITTO optimizes noise latents during inference to control text-to-music diffusion models, achieving superior performance in various tasks without fine-tuning.
Haojun Xia, Zhen Zheng, Xiaoxia Wu +10 authors
TC-FPx, a full-stack GPU kernel design, enables efficient six-bit quantization of large language models with Tensor Core support, achieving higher inference throughput on a single GPU.
Alexandre Ramé, Nino Vieillard, Léonard Hussenot +4 authors
Weight Averaged Reward Models (WARM) improve the alignment and quality of large language models by averaging fine-tuned reward models, addressing distribution shifts and preference inconsistencies.
Xinlei Chen, Zhuang Liu, Saining Xie +1 authors
A study on Denoising Diffusion Models reveals that only a few modern components are essential for self-supervised representation learning, leading to a simplified model resembling a Denoising Autoencoder.
Koichi Namekata, Amirmojtaba Sabour, Sanja Fidler +1 authors
Researchers develop a fine-grained image segmentor using Stable Diffusion without additional training, extracting semantic correspondences from low-dimensional feature maps for pixel-level segmentation.
Ka-Hei Hui, Aditya Sanghi, Arianna Rampini +4 authors
Make-A-Shape, a novel 3D generative model, efficiently trains using wavelet-tree representation and diffusion model techniques, generating high-quality shapes from various inputs in under two seconds.
Du Chen, Yi Huang, Xiaopu Li +7 authors
A multilingual large language model with 14 billion parameters achieves state-of-the-art performance across various tasks through data scheduling and fine-tuning.
Yiyuan Zhang, Xiaohan Ding, Kaixiong Gong +3 authors
Multimodal Pathway improves transformer models by incorporating relevant data from other modalities through a novel Cross-Modal Re-parameterization method, yielding consistent performance gains across various recognition tasks.
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