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。
Xin Chan, Xiaoyang Wang, Dian Yu +2 authors
Persona-driven synthetic data creation using a large language model and 1 billion diverse personas for various applications demonstrates versatility, scalability, and flexibility.
50 篇论文 · 按点赞排序
Pan Zhang, Xiaoyi Dong, Yuhang Zang +24 authors
A large-vision language model with long-contextual capabilities supports various comprehension and composition tasks, achieving GPT-4V level performance using 7B parameters.
Philippe Laban, Alexander R. Fabbri, Caiming Xiong +1 authors
The SummHay task evaluates LLMs and RAG systems on summarizing long-context documents by identifying relevant insights and citing sources, revealing challenges even for systems with document relevance signals.
Runqi Qiao, Qiuna Tan, Guanting Dong +15 authors
WE-MATH benchmark evaluates Large Multimodal Models in visual mathematical reasoning, focusing on knowledge acquisition and generalization principles beyond end-to-end performance.
Chunqiu Steven Xia, Yinlin Deng, Soren Dunn +1 authors
Agentless, a two-phase approach to software development automation, achieves high performance and low cost compared to complex autonomous agents on the SWE-bench Lite benchmark.
Junying Chen, Ruyi Ouyang, Anningzhe Gao +9 authors
The refinement of medical image-text pairs using multimodal large language models enhances medical multimodal capabilities and data quality, leading to superior performance in benchmarks.
Christopher E. Mower, Yuhui Wan, Hongzhan Yu +19 authors
A framework enables non-experts to program robots using natural language prompts and contextual information from the Robot Operating System (ROS), supported by large language models and offering various behavior modes, imitation learning, and feedback mechanisms.
Kepan Nan, Rui Xie, Penghao Zhou +6 authors
The paper introduces OpenVid-1M and OpenVidHD-0.4M datasets, along with MVDiT, a Multi-modal Video Diffusion Transformer, to enhance text-to-video generation by leveraging high-quality data and comprehensive text information.
Manuel Faysse, Hugues Sibille, Tony Wu +3 authors
The ColPali retrieval model uses Vision Language Models to create embeddings from images of document pages, improving performance and speed in visually rich document retrieval tasks.
Ivan Rubachev, Nikolay Kartashev, Yury Gorishniy +1 authors
TabReD is a new collection of industry-grade tabular datasets that address temporal dynamics and feature engineering, revealing that simpler models outperform more complex deep learning architectures in these settings.
Zihan Wang, Deli Chen, Damai Dai +3 authors
Research on parameter-efficient fine-tuning for Large Language Models with Mixture-of-Experts architecture reveals that task-specific expert tuning can enhance performance and efficiency.
Boyuan Chen, Diego Marti Monso, Yilun Du +3 authors
Diffusion Forcing trains diffusion models to denoise tokens with independent noise levels, combining next-token prediction and full-sequence diffusion strengths for sequence generation and decision-making tasks.
Qian Liu, Xiaosen Zheng, Niklas Muennighoff +5 authors
RegMix automatically identifies optimal data mixtures for pre-training large language models by formulating it as a regression task, significantly improving performance with reduced computational cost.
Ante Wang, Linfeng Song, Ye Tian +5 authors
A guided tree search algorithm with dynamic node selection and budget calculation improves LLM mathematical reasoning performance while reducing computational costs.
Jinsheng Huang, Liang Chen, Taian Guo +13 authors
MMEvalPro is a benchmark designed to mitigate biases in multimodal evaluations through augmented questions and rigorous metrics, demonstrating greater challenge and trustworthiness for LMM and LLM performance assessments.
Wael Mattar, Idan Levy, Nir Sharon +1 authors
Wavelet-based image generation uses a modified language transformer to model wavelet subband correlations for generating images conditioned on specific inputs.
Huiqiang Jiang, Yucheng Li, Chengruidong Zhang +9 authors
MInference, a sparse calculation method, accelerates pre-filling of long-context LLMs by leveraging specific patterns in attention matrices, reducing latency significantly without compromising accuracy.
Zimu Lu, Aojun Zhou, Ke Wang +4 authors
Step-Controlled DPO enhances large language models by providing targeted error supervision, improving reasoning and alignment compared to standard DPO.
Haofan Wang, Peng Xing, Renyuan Huang +3 authors
InstantStyle-Plus enhances style transfer by maintaining content integrity and semantic fidelity while efficiently integrating target styles using advanced components like ControlNet and semantic adapters.
Chang-Han Yeh, Chin-Yang Lin, Zhixiang Wang +3 authors
A zero-shot video restoration method using pre-trained image restoration diffusion models with hierarchical token merging and a hybrid correspondence mechanism achieves top performance and generalization across various datasets and degradations.
Seyedmorteza Sadat, Manuel Kansy, Otmar Hilliges +1 authors
Independent condition guidance and time-step guidance enhance diffusion model performance without special training or conditional information.
Elmira Amirloo, Jean-Philippe Fauconnier, Christoph Roesmann +8 authors
Preference alignment in Multimodal Large Language Models (MLLMs) is analyzed to understand which elements contribute to improved performance, leading to the introduction of Bias-Driven Hallucination Sampling (BDHS) for creating multimodal preference data.
Wentong Li, Yuqian Yuan, Jian Liu +4 authors
A novel visual projector using a coarse-to-fine scheme reduces visual token redundancy and improves visual reasoning in Multimodal LLMs.
Sefik Emre Eskimez, Xiaofei Wang, Manthan Thakker +10 authors
E2 TTS is a non-autoregressive zero-shot text-to-speech system using flow-matching-based mel spectrogram generation without additional components, achieving state-of-the-art performance.
Yixing Li, Yuxian Gu, Li Dong +3 authors
Direct Preference Knowledge Distillation (DPKD) improves the knowledge transfer from large language model teachers to students by incorporating an implicit reward function and distribution divergence.
Nataniel Ruiz, Yuanzhen Li, Neal Wadhwa +4 authors
Magic Insert allows style-aware drag-and-drop insertion of subjects from one image into another of a different style using domain adaptation and a fine-tuned diffusion model.
Zeyu Xie, Xuenan Xu, Zhizheng Wu +1 authors
PicoAudio, a temporal-controlled audio generation framework, enhances timestamp and occurrence frequency controllability through tailored model design and fine-grained audio-text data.
Xiaozhong Ji, Chuming Lin, Zhonggan Ding +7 authors
A new audio-driven face generation framework RealTalk uses an audio-to-expression transformer and a high-fidelity expression-to-face renderer for precise lip synchronization and high-quality real-time facial rendering.
Xiang Li, Cristina Mata, Jongwoo Park +8 authors
LLaRA, a framework for robotic action policies using conversation-style instruction, enhances performance through automatic data generation and fine-tuning of Vision Language Models.
Max Zuo, Francisco Piedrahita Velez, Xiaochen Li +2 authors
A benchmark evaluates language models' ability to generate semantically correct PDDL code from natural language descriptions, revealing significant gaps in current solutions.
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