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。
Aeree Cho, Grace C. Kim, Alexander Karpekov +5 authors
Transformer Explainer is an interactive visualization tool that allows non-experts to understand the inner workings of the GPT-2 model through real-time experimentation and visualization in a web browser.
50 篇论文 · 按点赞排序
Yuan Yao, Tianyu Yu, Ao Zhang +20 authors
MiniCPM-V presents a series of efficient Multimodal Large Language Models optimized for end-side deployment, offering high performance and practical usability compared to larger models.
Pengcheng Chen, Jin Ye, Guoan Wang +15 authors
GMAI-MMBench is a comprehensive general medical AI benchmark that evaluates large vision-language models across multiple medical domains and perceptual granularities, identifying key insufficiencies and highlighting areas for improvement.
Charlie Snell, Jaehoon Lee, Kelvin Xu +1 authors
Research on scaling test-time computation in LLMs finds that compute-optimal strategies can enhance performance and efficiency beyond larger base models.
Bo Li, Yuanhan Zhang, Dong Guo +7 authors
LLaVA-OneVision is a unified multimodal model that advances performance across single-image, multi-image, and video scenarios with strong transfer learning capabilities.
Fanqing Meng, Jin Wang, Chuanhao Li +9 authors
The Multimodal Multi-image Understanding (MMIU) benchmark assesses LVLMs across various multi-image tasks, highlighting challenges in spatial understanding.
Jiayuan Zhu, Yunli Qi, Junde Wu
MedSAM-2 extends the SAM 2 framework for 2D and 3D medical image segmentation, offering one-prompt segmentation across multiple images and outperforming existing models in various medical imaging tasks.
Xingguang Yan, Han-Hung Lee, Ziyu Wan +1 authors
Object Images convert 3D models into a manageable 2D format using image generation models like Diffusion Transformers, providing realistic 3D shapes and PBR materials.
Ziyang Ma, Yakun Song, Chenpeng Du +5 authors
A novel listening-while-speaking language model (LSLM) enhances real-time, full-duplex speech interaction by integrating speech generation and real-time audio input with an optimal fusion strategy.
Daniel Fleischer, Moshe Berchansky, Moshe Wasserblat +1 authors
RAG Foundry is an open-source framework integrating data creation, training, inference, and evaluation to facilitate the augmentation of large language models for RAG use cases, demonstrating improved performance across diverse datasets.
LG AI Research, Soyoung An, Kyunghoon Bae +35 authors
EXAONE 3.0, a 7.8B instruction-tuned language model from LG AI Research, demonstrates strong performance and instruction-following capability, particularly in Korean and across general complex reasoning tasks.
Dongyang Liu, Shitian Zhao, Le Zhuo +4 authors
Lumina-mGPT, a family of pretrained decoder-only transformer models, excels in various vision and language tasks, particularly text-to-image generation, through Flexible Progressive Supervised Finetuning and Ominiponent Supervised Finetuning.
Zaijing Li, Yuquan Xie, Rui Shao +3 authors
A multimodal agent, Optimus-1, equipped with a Hybrid Multimodal Memory module, significantly outperforms existing agents in long-horizon tasks by leveraging hierarchical knowledge graphs and abstracted multimodal experience pools.
Yunfei Xie, Ce Zhou, Lang Gao +8 authors
MedTrinity-25M, a large-scale multimodal medical dataset with extensive annotations, enables automated generation of detailed visual and textual descriptions, achieving state-of-the-art performance in tasks like VQA-RAD and PathVQA.
Yiwen Chen, Yikai Wang, Yihao Luo +5 authors
A new autoregressive transformer, MeshAnything V2, generates high-quality, artist-created meshes using a compact and efficient mesh tokenization method called Adjacent Mesh Tokenization (AMT).
David B. D'Ambrosio, Saminda Abeyruwan, Laura Graesser +24 authors
A robot agent achieves amateur human-level performance in competitive table tennis through a hierarchical policy architecture and real-time adaptation to unseen opponents.
Tianlu Wang, Ilia Kulikov, Olga Golovneva +7 authors
A self-improving evaluator trained with synthetic data outperforms traditional human-annotated evaluators and matches top-performing reward models on RewardBench.
Yingying Zhang, Xin Guo, Jiangwei Lao +7 authors
POA framework facilitates the pre-training of various-sized models through an innovative elastic student branch within self-distillation, achieving state-of-the-art performance across multiple backbones.
Yongzhi Xu, Yonhon Ng, Yifu Wang +5 authors
A novel deep-learning approach uses sketch-based input and pre-trained models to generate interactive 3D game scenes with accurate layouts.
Mervat Abassy, Kareem Elozeiri, Alexander Aziz +21 authors
LLM-DetectAIve is a system that classifies texts as human-written, machine-generated, machine-written machine-humanized, and human-written machine-polished to enhance integrity in domains like education and academia.
Ciara Rowles, Shimon Vainer, Dante De Nigris +3 authors
IPAdapter-Instruct integrates natural-image conditioning with instruction prompts to efficiently handle multiple tasks in image generation without significant quality degradation.
François Remy, Pieter Delobelle, Hayastan Avetisyan +3 authors
A novel trans-tokenization strategy is introduced to transfer vocabulary from high-resource languages to low-resource languages, enabling competitive performance in various tasks and facilitating zero-shot machine translation for languages like Tatar.
Riccardo Orlando, Pere-Lluis Huguet-Cabot, Edoardo Barba +1 authors
ReLiK, a Retriever-Reader architecture, achieves state-of-the-art performance in Entity Linking and Relation Extraction while significantly reducing inference speed.
Prannaya Gupta, Le Qi Yau, Hao Han Low +8 authors
WalledEval is a comprehensive safety testing toolkit for large language models using multiple safety benchmarks and custom mutators to assess performance in areas such as multilingual safety, exaggerated safety, and cultural contexts.
Xiangyan Liu, Bo Lan, Zhiyuan Hu +5 authors
A system integrating LLMs with graph database interfaces for improved codebase interaction demonstrates competitive performance in real-world coding applications.
Yulei Qin, Yuncheng Yang, Pengcheng Guo +7 authors
A review of existing data assessment and selection methods for instruction tuning of large language models categorizes them into quality-based, diversity-based, and importance-based approaches, discusses their limitations, and outlines future research directions.
Thao Nguyen, Jeffrey Li, Sewoong Oh +4 authors
The proposed instruction back-and-forth translation method enhances synthetic data for large language models, improving win rates and quality compared to existing datasets.
Alon Jacovi, Moran Ambar, Eyal Ben-David +5 authors
CoverBench is a challenging benchmark for verifying the correctness of language model outputs in complex reasoning settings, providing diverse evaluations across domains and reasoning types.
Ioannis Siglidis, Aleksander Holynski, Alexei A. Efros +2 authors
The approach uses fine-tuned conditional diffusion models to define a typicality measure for visual data elements across various labels and datasets, improving scalability and versatility in visual data mining.
Qirui Jiao, Daoyuan Chen, Yilun Huang +2 authors
A novel dataset, Img-Diff, enhances fine-grained image recognition in MLLMs using contrastive learning and image difference captioning, improving performance over larger datasets.
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