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。
Raymond Li, Loubna Ben Allal, Yangtian Zi +64 authors
StarCoder, a 15.5B parameter LLM trained on 1 trillion tokens, outperforms other open Code LLMs across multiple languages and fine-tuned Python, with safety enhancements and publicly available under the Open Responsible AI Model license.
50 篇论文 · 按点赞排序
Jinyang Li, Binyuan Hui, Ge Qu +13 authors
Bird, a large-scale text-to-SQL benchmark, addresses challenges posed by big databases, emphasizing database values, external knowledge, and SQL efficiency.
Shashank Rajput, Nikhil Mehta, Anima Singh +10 authors
A generative retrieval model using Semantic IDs and a Transformer sequence-to-sequence approach improves recommendation system performance and generalization, especially for cold-start items.
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.
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.
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.
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.
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 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.
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.
Xinyu Liu, Houwen Peng, Ningxin Zheng +3 authors
EfficientViT addresses memory inefficiency and computational redundancy in existing vision transformers, offering improved speed and accuracy.
Xianzhi Li, Xiaodan Zhu, Zhiqiang Ma +2 authors
ChatGPT and GPT-4 perform well on numerical reasoning tasks in financial texts but struggle with financial NER and sentiment analysis, compared to domain-specific models.
Zhaoyang Liu, Yinan He, Wenhai Wang +16 authors
The InternChat framework combines chatbots with pointing gestures to enhance the precision and efficiency of vision-centric tasks, particularly improving the accuracy of large vision-language models in complex visual scenarios.
Arijit Ray, Filip Radenovic, Abhimanyu Dubey +3 authors
A novel text-to-image retrieval benchmark (Cola) is used to assess and enhance compositional reasoning in vision-language models through fine-tuning strategies.
Jiacheng Liu, Wenya Wang, Dianzhuo Wang +3 authors
Vera, a model trained on commonsense statements, effectively verifies and separates correct from incorrect statements across diverse domains, outperforming existing models and aiding in detecting erroneous statements from language models.
Shanshan Zhong, Zhongzhan Huang, Wushao Wen +2 authors
A parameter-efficient fine-tuning method, SUR-adapter, is proposed to enhance the semantic understanding and reasoning of diffusion models for text-to-image generation, using knowledge distillation from large language models.
Xianhang Li, Zeyu Wang, Cihang Xie
Reducing the token length in CLIP training improves scaling efficiency, allowing academic researchers to train high-performing models with limited resources.
KunChang Li, Yinan He, Yi Wang +6 authors
VideoChat combines video foundation models and large language models with a learnable neural interface for tasks such as spatiotemporal reasoning and causal relationship inference, supported by a new video-centric instruction dataset.
Pradyumna Reddy, Scott Wisdom, Klaus Greff +2 authors
AudioSlots, a slot-centric generative model based on permutation-equivariant networks, demonstrates potential for blind source separation using transformed audio spectrograms.
Lei Wang, Wanyu Xu, Yihuai Lan +4 authors
A new prompting strategy, Plan-and-Solve (PS) and its extension PS+, improves the reasoning capabilities of LLMs by reducing errors in multi-step reasoning tasks, outperforming existing zero-shot methods.
David Futschik, Kelvin Ritland, James Vecore +5 authors
A novel learning-based method improves lighting in portraits by softening shadows and specular highlights while preserving overall illumination, and enhances higher-level vision applications.
Hongyuan Lu, Haoyang Huang, Dongdong Zhang +3 authors
CoD enhances LLMs in MNMT by incorporating multilingual dictionaries, significantly improving translation performance, especially for low-resource languages.
Hong Wang, Xuan Luo, Weizhi Wang +1 authors
FLAIR framework detects conversational bots online using targeted questions to differentiate between human and artificial responses.
Chenghao Li, Chaoning Zhang, Atish Waghwase +5 authors
This survey compiles and explains the advancements in text-to-3D generation, covering foundational technologies, data representations, and applications across various domains.
Rohan Dhesikan, Vignesh Rajmohan
A novel method combines zero-shot text-to-video generation with ControlNet for improved video content creation, demonstrating high-quality and consistent video outputs.
Foivos Paraperas Papantoniou, Alexandros Lattas, Stylianos Moschoglou +1 authors
A novel approach uses diffusion models to reconstruct 3D facial BRDF from a single image by sampling and inpainting UV textures.
Ruilong Li, Hang Gao, Matthew Tancik +1 authors
NerfAcc accelerates Neural Radiance Field training and rendering by providing advanced, flexible sampling methods that reduce training time significantly.
Jimmy Wu, Rika Antonova, Adam Kan +6 authors
Robots can learn user preferences for household cleanup using few-shot language-based planning and perception with large language models, achieving high accuracy in both simulated and real-world scenarios.
Chenxiao Liu, Shuai Lu, Weizhu Chen +5 authors
CodeExecutor, a Transformer model enhanced with code execution pre-training and curriculum learning, demonstrates improved performance in code execution tasks and benefits for code intelligence tasks like zero-shot search and text-to-code generation.
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