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。
Yichun Shi, Peng Wang, Jianglong Ye +3 authors
MVDream generates geometrically consistent multi-view images from text prompts using pre-trained image diffusion models and Score Distillation Sampling, improving 3D generation stability and supporting personalized generation.
22 篇论文 · 按点赞排序
Lukas Blecher, Guillem Cucurull, Thomas Scialom +1 authors
Nougat, a Visual Transformer model, performs OCR on scientific documents converting them to markup language, enhancing digital accessibility.
Chi Han, Qifan Wang, Wenhan Xiong +3 authors
LM-Infinite addresses length generalization in Transformer-based LLMs using a simple $\Lambda$-shaped attention mask, achieving fluency and quality up to 32k tokens without retraining.
Yu Shu, Siwei Dong, Guangyao Chen +5 authors
LLaSM, an end-to-end trained large multi-modal speech-language model, enhances human interaction with AI by following speech-and-language instructions using cross-modal conversational abilities.
Scott L. Fleming, Alejandro Lozano, William J. Haberkorn +27 authors
MedAlign, a curated dataset of natural language instructions for EHR data, evaluates the accuracy and quality of general domain LLMs in healthcare, revealing high error rates and context length dependencies.
Neha Sengupta, Sunil Kumar Sahu, Bokang Jia +19 authors
Jais and Jais-chat, Arabic-centric LLMs based on the GPT-3 architecture, outperform existing models in Arabic and are competitive in English with fewer resources.
Wenqi Shao, Mengzhao Chen, Zhaoyang Zhang +7 authors
OmniQuant improves LLM deployment by optimizing quantization parameters through Learnable Weight Clipping and Learnable Equivalent Transformation, enhancing performance across various quantization settings.
Yaodong Yu, Tianzhe Chu, Shengbang Tong +4 authors
Transformers designed with an emphasis on low-dimensional structures, such as CRATE, can develop segmentation properties even with minimal supervised training, challenging the notion that such properties require intricate self-supervised mechanisms.
Ensheng Shi, Fengji Zhang, Yanlin Wang +6 authors
SoTaNa utilizes ChatGPT-generated data and parameter-efficient fine-tuning to enhance LLaMA, providing an accessible open-source software development assistant capable of answering Stack Overflow questions, code summarization, and generation.
Xiangru Tang, Bill Qian, Rick Gao +3 authors
BioCoder evaluates pre-trained models for bioinformatics code generation, focusing on domain knowledge and complex functional dependencies.
Mel Vecerik, Carl Doersch, Yi Yang +6 authors
Robots can learn new tasks more quickly and generally using dense tracking and parameterized low-level controllers, as demonstrated with TAP models.
Qingping Zheng, Yuanfan Guo, Jiankang Deng +4 authors
The Any-Size-Diffusion model addresses resolution issues in text-to-image synthesis by using a two-stage pipeline that optimizes text-conditional diffusion and allows for rapid enlargement of images to any size without artifacts.
Daoguang Zan, Ailun Yu, Bo Shen +8 authors
Programming languages enhance each other during the instruction fine-tuning phase of code large language models, as demonstrated by significant improvements in test performance across different languages.
Zike Yan, Haoxiang Yang, Hongbin Zha
The paper introduces Active Neural Mapping, a method using a coordinate-based implicit neural representation to guide an agent for active exploration and mapping in unseen environments, leveraging the uncertainty in the neural field weights and geometric information.
Lucas Bandarkar, Davis Liang, Benjamin Muller +7 authors
Belebele, a multilingual MRC dataset, evaluates the performance of LLMs and MLMs across numerous language variants, highlighting the effectiveness of models trained on balanced multilingual data.
Yanjie Ze, Ge Yan, Yueh-Hua Wu +6 authors
GNFactor, a visual behavior cloning agent for multi-task robotic manipulation, uses a generalizable neural field and a Perceiver Transformer to achieve high generalization on seen and unseen tasks.
Stephan Rasp, Stephan Hoyer, Alexander Merose +15 authors
WeatherBench 2 provides an improved framework for evaluating data-driven weather forecasting models using established operational metrics and an open-source dataset.
Matthew Dutson, Yin Li, Mohit Gupta
The proposed Eventful Transformers identify and reprocess only significant changes in video frames to reduce computational costs by 2-4x with minimal accuracy loss.
Jiaben Chen, Huaizu Jiang
The SportsSloMo benchmark improves human-centric video frame interpolation through human-aware loss terms, enhancing performance in sports slow-motion video generation.
Chong Zeng, Guojun Chen, Yue Dong +3 authors
A new neural implicit radiance representation models object relighting from a small set of photographs using signed distance functions and two MLPs, incorporating shadow and highlight hints for high-frequency light transport effects.
Andrea Bajcsy, Antonio Loquercio, Ashish Kumar +1 authors
A fully-observable robot policy guides a partially-observable one in pursuit-evasion tasks by generating supervision, focusing on evader behavior diversity and modeling assumptions.
Minheng Ni, Chenfei Wu, Xiaodong Wang +4 authors
The Two-stage Intervention (TIN) framework uses a large-scale language model and diffusion synthesis model to enable responsible visual synthesis by avoiding forbidden visual concepts while following user queries.
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