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。
47 篇论文 · 按点赞排序
Ziyang Yuan, Mingdeng Cao, Xintao Wang +3 authors
CustomNet enhances text-to-image generation by integrating 3D novel view synthesis to customize object viewpoints, locations, and backgrounds, ensuring identity preservation and diverse outputs without test-time optimization.
Ernie Chang, Pin-Jie Lin, Yang Li +6 authors
A retrieval-based framework enhances audio quality in text-to-audio generation by revisiting and editing user prompts with reference to training exemplars.
Yichi Zhang, Jiayi Pan, Yuchen Zhou +2 authors
Visual-Language Models exhibit susceptibility to visual illusions, similar to humans, with larger models showing higher alignment and susceptibility compared to smaller models.
Qiaoling Chen, Qinghao Hu, Zhisheng Ye +4 authors
A novel LLM training framework, AMSP, addresses scaling challenges for large-scale token processing using model state partitioning and a scale-aware partitioner.
Mingjie Liu, Teo Ene, Robert Kirby +33 authors
Custom tokenizers, continued pretraining, supervised fine-tuning, and retrieval models improve large language models for industrial chip design tasks, including engineering assistance, EDA script generation, and bug analysis.
Nan He, Hanyu Lai, Chenyang Zhao +12 authors
TeacherLM-7.1B significantly enhances NLP tasks through zero-shot performance and data augmentation, surpassing larger models and improving various student models.
Pierre Sermanet, Tianli Ding, Jeffrey Zhao +18 authors
A scalable data collection scheme using both robots and humans within office buildings enhances the performance of models in robotics-focused visual question answering, particularly for long-horizon tasks, and demonstrates the importance of grounded data for real-world deployment.
Tian Liang, Zhiwei He, Jen-tes Huang +7 authors
A new evaluation framework using interactive multi-agent games assesses LLMs' intelligence, expression, and strategic communication skills.
Zhiling Yan, Kai Zhang, Rong Zhou +3 authors
GPT-4V's evaluation on medical VQA using multi-modal datasets shows unreliability and highlights several limitations.
Hailin Chen, Amrita Saha, Steven Hoi +1 authors
A personalized distillation process improves smaller open-sourced LLMs by focusing on mistakes made by the student model, outperforming standard distillation methods with limited data.
Kyle Sargent, Zizhang Li, Tanmay Shah +8 authors
A 3D-aware diffusion model, ZeroNVS, addresses challenges in single-image novel view synthesis for complex, in-the-wild scenes, setting new benchmarks and improving diversity in synthesized views.
Ahmed Magooda, Alec Helyar, Kyle Jackson +14 authors
A framework measures responsible AI metrics for large language models to assess and mitigate harm, leveraging existing expertise and advanced language model capabilities.
Tristan Aumentado-Armstrong, Ashkan Mirzaei, Marcus A. Brubaker +4 authors
Combining autoencoders with Neural Radiance Fields improves rendering speed and image quality, enabling efficient and high-fidelity 3D scene representation suitable for robotics.
Xin Yu, Yuan-Chen Guo, Yangguang Li +3 authors
Classifier Score Distillation leverages classifier-free guidance as a standalone method for effective text-to-3D generation tasks, outperforming state-of-the-art techniques.
Tianwen Wei, Liang Zhao, Lichang Zhang +27 authors
Skywork-13B, a bilingual LLM trained on extensive corpora, excels in benchmarks and achieves state-of-the-art performance in Chinese language modeling, introducing a novel leakage detection method.
Andrew Szot, Max Schwarzer, Harsh Agrawal +6 authors
LLaRP, a reinforcement learning approach adapting pre-trained large language models for embodied visual tasks, achieves high success rates on unseen tasks and generalizes robustly to new instructions.
Nitish Joishi, Javier Rando, Abulhair Saparov +2 authors
Language models can identify truthfulness in text by modeling a truthful persona shared by reliable sources, allowing them to infer truth across different contexts and agents.
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