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。
50 篇论文 · 按点赞排序
Yao Zhao, Rishabh Joshi, Tianqi Liu +3 authors
Sequence Likelihood Calibration (SLiC) is shown to be an effective and simpler alternative to Reinforcement Learning from Human Feedback (RLHF) for learning from human preferences in language models.
Yue Wang, Hung Le, Akhilesh Deepak Gotmare +3 authors
CodeT5+, a family of flexible encoder-decoder LLMs initialized with frozen LLMs and trained with a mixture of pretraining objectives, achieves state-of-the-art performance across various code-related tasks.
Zhen Ye, Wei Xue, Xu Tan +3 authors
CoMoSpeech, a consistency model-based method, achieves high-quality speech synthesis with a single diffusion sampling step, significantly speeding up inference compared to traditional models.
Wenliang Dai, Junnan Li, Dongxu Li +6 authors
InstructBLIP models, built through instruction-aware visual feature extraction on BLIP-2, achieve top performance in vision-language tasks, including zero-shot evaluation and fine-tuned downstream tasks.
Lingjiao Chen, Matei Zaharia, James Zou
Strategies including prompt adaptation, LLM approximation, and LLM cascade are proposed to reduce costs and maintain or improve performance when using large language models.
Bo Li, Yuanhan Zhang, Liangyu Chen +3 authors
Multimodal instruction tuning improves the performance of models like Otter, based on OpenFlamingo, with an optimized dataset and reduced training resource requirements.
Chen-Yu Lee, Chun-Liang Li, Hao Zhang +13 authors
FormNetV2 uses a centralized multimodal graph contrastive learning approach to improve self-supervised pre-training and achieves state-of-the-art performance on form understanding benchmarks with a compact model.
Ravid Shwartz-Ziv, Yann LeCun
A unified framework integrating information theory to formalize the self-supervised learning problem is presented, exploring the intersection with deep neural networks.
Yuma Koizumi, Heiga Zen, Shigeki Karita +7 authors
A new speech dataset called LibriTTS-R, derived by applying speech restoration to LibriTTS, improves sound quality and enhances neural end-to-end TTS performance.
Yuchao Gu, Xintao Wang, Jay Zhangjie Wu +10 authors
Mix-of-Show addresses challenges in decentralized multi-concept customization for text-to-image diffusion models using embedding-decomposed LoRA, gradient fusion, and regionally controllable sampling.
Paul S. Scotti, Atmadeep Banerjee, Jimmie Goode +9 authors
MindEye, a novel fMRI-to-image approach, retrieves and reconstructs viewed images with high accuracy using specialized submodules and diffusion prior in latent space.
Arnav Gudibande, Eric Wallace, Charlie Snell +5 authors
Finetuning open-source language models on outputs from proprietary models like ChatGPT shows improved instruction-following but fails to close the performance gap on unsupported tasks.
Liying Cheng, Xingxuan Li, Lidong Bing
GPT-4 demonstrates comparable performance to human data analysts in end-to-end data analysis tasks across various domains.
Tiedong Liu, Bryan Kian Hsiang Low
A fine-tuned LLaMA model, Goat, outperforms GPT-4 in arithmetic tasks through supervised learning and task decomposition.
Hiroki Furuta, Ofir Nachum, Kuang-Huei Lee +3 authors
An instruction-following multimodal agent named WebGUM, trained offline using vision transformers and language models, significantly outperforms existing methods in web navigation and multi-step reasoning tasks.
Karan Singhal, Tao Tu, Juraj Gottweis +28 authors
Med-PaLM 2, built on PaLM 2 and enhanced with medical fine-tuning and ensemble refinement, significantly improves physician-level performance in medical question answering by scoring 86.5% on MedQA and excelling on other clinical datasets.
James Betker
TorToise is an expressive multi-voice speech synthesis system that applies autoregressive transformers and DDPMs from image generation to text-to-speech.
Jianyi Wang, Zongsheng Yue, Shangchen Zhou +2 authors
A novel method leverages pre-trained text-to-image diffusion models for blind super-resolution, using a time-aware encoder and a controllable feature wrapping module to enhance fidelity and adaptability to varying resolutions.
Dahun Kim, Anelia Angelova, Weicheng Kuo
Region-aware Open-vocabulary Vision Transformers improve open-vocabulary object detection and image-text retrieval by using focal loss and novel object proposals in contrastive learning.
Jianyi Zhang, Saeed Vahidian, Martin Kuo +4 authors
Federated Instruction Tuning (FedIT) leverages federated learning to improve generative large language models with diverse local instructions while maintaining privacy.
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