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。
Hugo Touvron, Louis Martin, Kevin Stone +65 authors
Llama 2, a series of pretrained and fine-tuned large language models, achieves superior performance in dialogue tasks compared to open-source alternatives and offers safety enhancements.
42 篇论文 · 按点赞排序
Yutao Sun, Li Dong, Shaohan Huang +5 authors
Retentive Network (RetNet) offers a foundation for large language models with parallel training, low-cost inference, and efficient long-sequence modeling through a retention mechanism supporting parallel, recurrent, and chunkwise recurrent paradigms.
Michal Geyer, Omer Bar-Tal, Shai Bagon +1 authors
A framework for text-driven video editing uses diffusion models to ensure consistency in the edited video by propagating features based on inter-frame correspondences.
Jean Kaddour, Joshua Harris, Maximilian Mozes +3 authors
Large Language Models (LLMs) went from non-existent to ubiquitous in the machine learning discourse within a few years. Due to the fast pace of the field, it is difficult to identify the remaining challenges and already fruitful application areas. In this paper, we aim to establish a systematic set of open problems and application successes so that ML researchers can comprehend the field's current state more quickly and become productive.
Liangyu Zha, Junlin Zhou, Liyao Li +22 authors
TableGPT, a fine-tuned language model with global tabular representations, enables seamless interaction and complex operations on tables via natural language and functional commands.
Sarah Gao, Andrew Kean Gao
A tool called Constellation clusters and visualizes large language models uploaded to Hugging Face using hierarchical clustering and n-gram analysis.
Yiyuan Zhang, Kaixiong Gong, Kaipeng Zhang +4 authors
Meta-Transformer unifies multimodal learning across 12 unpaired modalities using a frozen encoder in a shared token space.
Timo I. Denk, Yu Takagi, Takuya Matsuyama +4 authors
A method reconstructs music from fMRI data using the MusicLM model, matching semantic properties of original stimuli and identifying brain regions involved in processing music descriptions.
Tian Lan, Deng Cai, Yan Wang +2 authors
Text generation is improved by copying text segments from a pre-existing collection, resulting in better quality and efficiency compared to traditional methods.
Dimitri von Rütte, Elisabetta Fedele, Jonathan Thomm +1 authors
FABRIC is a training-free method that uses feedback images and self-attention layers to enhance diffusion-based text-to-image models through iterative human feedback.
Yang Zhao, Zhijie Lin, Daquan Zhou +3 authors
BuboGPT is a multi-modal LLM with visual grounding that enhances understanding and interaction between vision, audio, and language.
Ziyue Jiang, Jinglin Liu, Yi Ren +8 authors
Mega-TTS 2 extends zero-shot text-to-speech with arbitrary-length prompts by incorporating a multi-reference timbre encoder and a prosody language model, enhancing both identity preservation and prosodic control.
Lingjiao Chen, Matei Zaharia, James Zou
The performance and behavior of GPT-3.5 and GPT-4 fluctuated significantly between March and June 2023 across various tasks, emphasizing the necessity for ongoing LLM quality monitoring.
Lichang Chen, Shiyang Li, Jun Yan +8 authors
A data selection strategy using a strong LLM improves instruction-following models by filtering high-quality data, enhancing performance and reducing training time.
Liang Wang, Nan Yang, Furu Wei
A framework is proposed for training dense retrievers to select high-quality in-context examples for large language models, enhancing their performance across various tasks.
Daiqing Li, Huan Ling, Amlan Kar +5 authors
DreamTeacher uses generative networks to distill knowledge into self-supervised image backbones, improving performance across various benchmarks and datasets without labeled data.
Soumik Mukhopadhyay, Matthew Gwilliam, Vatsal Agarwal +5 authors
Diffusion models trained on image generation tasks produce useful feature embeddings that perform well for classification tasks and outperform generative-discriminative models like BigBiGAN.
Nikita Karaev, Ignacio Rocco, Benjamin Graham +3 authors
CoTracker, a deep learning architecture combining transformer networks and specialized attention layers, jointly tracks multiple video points across frames, offering superior performance compared to existing methods.
Bo-Ru Lu, Nikita Haduong, Chia-Hsuan Lee +7 authors
DIALGEN, a human-in-the-loop framework using ChatGPT, improves structured summarization of agent-client conversations by generating and refining subdialogues with human feedback.
Xinyang Zhang, Wentian Zhao, Xin Lu +1 authors
A method using autoencoders and diffusion models generates high-quality layered images, improving compositing workflows and mask quality.
Seonghyeon Ye, Doyoung Kim, Sungdong Kim +6 authors
FLASK evaluates large language models using a fine-grained protocol that assesses skills required for following user instructions across different domains and difficulty levels.
Norman Di Palo, Arunkumar Byravan, Leonard Hasenclever +3 authors
A framework using language as a reasoning tool in RL agents improves exploration efficiency and data reuse, and enhances the acquisition of skills for novel tasks.
Jianguo Zhang, Kun Qian, Zhiwei Liu +7 authors
DialogStudio is a comprehensive and diverse collection of dialogue datasets designed to address challenges in conversational AI and improve dialogue research and model training.
Thao Nguyen, Samir Yitzhak Gadre, Gabriel Ilharco +2 authors
Generated captions improve the utility of web-scraped image-text datasets by reducing noise without compromising diversity, outperforming existing filtering methods across multiple benchmarks and tasks.
Christopher Rawles, Alice Li, Daniel Rodriguez +2 authors
A dataset named Android in the Wild challenges device-control systems with natural language instructions, requiring semantic understanding and inference of precise gestures from visual appearance.
Tom Lieberum, Matthew Rahtz, János Kramár +3 authors
Circuit analysis applied to a 70B Chinchilla model provides insights into the mechanisms of multiple-choice question answering, demonstrating that logit attribution, attention pattern visualization, and activation patching scale and allow identification of output nodes like attention heads and MLPs.
Yuying Ge, Yixiao Ge, Ziyun Zeng +2 authors
SEED, an image tokenizer, enables Large Language Models to perform image-to-text and text-to-image generation by incorporating independent 1D image tokens optimized for semantic alignment and reconstruction.
Raphael Boige, Yannis Flet-Berliac, Arthur Flajolet +2 authors
Pretrained Action-State Transformer Agents (PASTA) are investigated for a range of reinforcement learning tasks using unified methodologies and parameter-efficient fine-tuning.
Tongshuang Wu, Haiyi Zhu, Maya Albayrak +21 authors
Modern LLMs can simulate some aspects of complex crowdsourcing pipelines but their success varies based on task specifics, instruction clarity, and interaction modality.
Jiahui Huang, Leonid Sigal, Kwang Moo Yi +2 authors
Interactive Neural Video Editing (INVE) improves real-time video editing by leveraging hash-grids encoding and bi-directional functions, enabling faster processing and a wider range of edit operations compared to Layered Neural Atlas (LNA).
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