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。
Chameleon Team
Chameleon is a mixed-modal early-fusion token-based model that achieves state-of-the-art performance across various tasks, including image captioning, text generation, and long-form mixed-modal generation, using a unified architecture.
29 篇论文 · 按点赞排序
Dan Biderman, Jose Gonzalez Ortiz, Jacob Portes +9 authors
LoRA, a parameter-efficient finetuning method for large language models, underperforms full finetuning in target domains but provides better regularization and maintains diverse generation compared to other techniques.
Hugo Laurençon, Léo Tronchon, Matthieu Cord +1 authors
Idefics2, a vision-language model with 8 billion parameters, achieves state-of-the-art performance on multimodal benchmarks through extensive experimental validation.
Hanze Dong, Wei Xiong, Bo Pang +7 authors
Online iterative reinforcement learning from human feedback achieves state-of-the-art performance in large language models using open-source datasets and proxy preference models.
Ruiqi Gao, Aleksander Holynski, Philipp Henzler +5 authors
CAT3D generates consistent 3D views from limited input using a multi-view diffusion model, enabling fast and accurate 3D reconstruction.
Abhijit Bendale, Michael Sapienza, Steven Ripplinger +3 authors
SUTRA, a multilingual Large Language Model architecture, achieves superior performance on multilingual tasks by decoupling conceptual understanding from language-specific processing using a Mixture of Experts framework.
Xueyan Niu, Bo Bai, Lei Deng +1 authors
A theoretical framework using Hopfield networks explains memorization and performance dynamics in Transformer-based language models, showing bounds on achievable cross-entropy loss.
Tianhe Ren, Qing Jiang, Shilong Liu +13 authors
Grounding DINO 1.5, comprising Pro and Edge models, enhances open-set object detection through advanced architectures and large datasets, achieving high performance and efficiency.
Yixing Jiang, Jeremy Irvin, Ji Hun Wang +3 authors
Multimodal foundation models exhibit significant performance improvements with many-shot in-context learning compared to few-shot, with Gemini 1.5 Pro showing higher data efficiency and better batch processing capabilities across various domains.
Raghu Prabhakar, Ram Sivaramakrishnan, Darshan Gandhi +27 authors
Samba-CoE, a Composition of Experts system with a three-tier memory architecture, achieves significant speedups and efficiency in deploying large language models using SambaNova's Reconfigurable Dataflow Unit.
Siwei Wang, Yifei Shen, Shi Feng +3 authors
Theory and experiments show that Transformer-based models, while capable of learning path-finding through embedding adjacency matrices, struggle with identifying reachability through transitivity in network planning tasks.
Wenqi Dong, Bangbang Yang, Lin Ma +5 authors
Coin3D is an interactive 3D modeling framework that uses a coarse geometry proxy to control 3D generation, supports seamless local editing, and provides responsive previews using diffusion models and various novel techniques.
Ethan Weber, Riley Peterlinz, Rohan Mathur +3 authors
A pipeline combining camera pose estimation and image deformation corrects 2D drawing inconsistencies in hand-drawn cartoons to recover plausible 3D structures for novel-view synthesis.
Wanting Xu, Yang Liu, Langping He +2 authors
Xmodel-VLM is a lightweight multimodal vision language model developed for efficient deployment, achieving performance comparable to larger models with a 1B language model trained using the LLaVA paradigm.
Zhimin Li, Jianwei Zhang, Qin Lin +42 authors
Hunyuan-DiT, a text-to-image diffusion transformer, achieves state-of-the-art in Chinese-to-image generation by incorporating fine-grained language understanding and multi-turn dialogue capabilities.
Qi Chen, Xiubo Geng, Corby Rosset +28 authors
MS MARCO Web Search provides a large-scale, information-rich web dataset with real-world query-document labels for training and evaluating neural indexer models, embedding models, and information access systems.
Xinyang Li, Zhangyu Lai, Linning Xu +5 authors
Dual3D is a text-to-3D generation framework that combines dual-mode latent diffusion models with neural surfaces and pre-trained components for rapid and high-quality 3D asset creation.
Junqin Huang, Zhongjie Hu, Zihao Jing +2 authors
Piccolo2, an embedding model, achieves new state-of-the-art performance across six tasks using a multi-task hybrid loss approach, scaled embedding dimensions, and MRL training.
Vivian Liu, Rubaiat Habib Kazi, Li-Yi Wei +4 authors
LogoMotion is an LLM-based system that generates animation code for logos, producing visually-grounded and content-aware animations.
Chengyue Wu, Yixiao Ge, Qiushan Guo +5 authors
Plot2Code is a visual coding benchmark for evaluating Multi-modal Large Language Models' abilities to convert visual figures into executable code, using automatic metrics for comprehensive assessment.
Nilaksh Das, Saket Dingliwal, Srikanth Ronanki +13 authors
SpeechVerse, a multi-task training framework, enhances performance on diverse speech processing tasks with natural language instructions by combining pre-trained speech and text models.
Yunhao Tang, Daniel Zhaohan Guo, Zeyu Zheng +8 authors
Offline reinforcement learning from human feedback underperforms online methods in large language model alignment, suggesting that on-policy sampling is crucial and highlighting challenges in offline alignment.
Yingjie Zhai, Wenshuo Li, Yehui Tang +2 authors
The proposed SqueezeTime network enhances video understanding on mobile devices by integrating the temporal axis into the channel dimension, improving accuracy and GPU throughput.
Weili Nie, Sifei Liu, Morteza Mardani +3 authors
BlobGEN, a blob-grounded text-to-image diffusion model with masked cross-attention and in-context learning from LLMs, excels in zero-shot generation and layout control, outperforming benchmarks in numerical and spatial correctness.
Yunhao Ge, Yihe Tang, Jiashu Xu +20 authors
The BEHAVIOR Vision Suite provides tools and assets for generating customized synthetic data to evaluate computer vision models systematically under varying conditions.
Yunfan Jiang, Chen Wang, Ruohan Zhang +2 authors
TRANSIC integrates simulation policies with human corrections to achieve successful sim-to-real transfer in complex robotic tasks.
Raghuveer Peri, Sai Muralidhar Jayanthi, Srikanth Ronanki +11 authors
Research investigates adversarial vulnerabilities and proposes countermeasures for speech-language models that follow speech instructions and generate text responses.
James Oswald, Kavitha Srinivas, Harsha Kokel +3 authors
Large language models show moderate proficiency in generating automated planning domain models from natural language descriptions.
Emilian Postolache, Natalia Polouliakh, Hiroaki Kitano +3 authors
Latent diffusion models are employed to reconstruct naturalistic music from EEG recordings using an end-to-end training approach and evaluating with neural embedding-based metrics.
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