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。
Chris Lu, Cong Lu, Robert Tjarko Lange +3 authors
The AI Scientist is a framework that enables independent scientific research through automatic idea generation, experimentation, and paper writing using large language models.
50 篇论文 · 按点赞排序
Zhenting Qi, Mingyuan Ma, Jiahang Xu +3 authors
rStar enhances small language models' reasoning capabilities through a self-play mutual generation-discrimination process without fine-tuning, improving accuracy across various reasoning tasks.
Yushi Bai, Jiajie Zhang, Xin Lv +6 authors
AgentWrite enables long context LLMs to generate outputs exceeding 20,000 words by decomposing tasks into subtasks, using a new dataset to train models with extended output capabilities.
Huajian Xin, Z. Z. Ren, Junxiao Song +14 authors
DeepSeek-Prover-V1.5 improves theorem proving by optimizing training and inference, utilizing reinforcement learning, and proposing RMaxTS for diverse proof paths, achieving state-of-the-art results on miniF2F and ProofNet benchmarks.
Bohao Peng, Jian Wang, Yuechen Zhang +3 authors
ControlNeXt is a computationally efficient method for controllable image and video generation with a streamlined architecture and minimal parameters, using Cross Normalization for fast training convergence.
Clément Christophe, Praveen K Kanithi, Tathagata Raha +2 authors
Med42-v2 enhances Llama3 with clinical data to improve performance in healthcare settings, outperforming generic models across medical benchmarks.
Chaoyou Fu, Haojia Lin, Zuwei Long +12 authors
VITA, an open-source Multimodal Large Language Model, excels in processing Video, Image, Text, and Audio with seamless interaction, showcasing advancements in multimodal understanding and human-computer interaction.
Tom Lieberum, Senthooran Rajamanoharan, Arthur Conmy +7 authors
Gemma Scope is an open suite of JumpReLU SAEs trained on various layers of Gemma 2 models to facilitate safety and interpretability research.
Kexun Zhang, Weiran Yao, Zuxin Liu +13 authors
A DEI framework enhances software engineering problem-solving by leveraging a collective of LLM agents, achieving higher resolve rates compared to individual agents and closed-source solutions.
Zhuoyi Yang, Jiayan Teng, Wendi Zheng +16 authors
CogVideoX is a large-scale diffusion transformer model using a 3D Variational Autoencoder and expert transformer for generating high-quality, coherent videos from text prompts.
Yiming Liang, Ge Zhang, Xingwei Qu +9 authors
I-SHEEP, an iterative self-enhancement paradigm, enables LLMs to continuously self-align from scratch, achieving significant improvements across various benchmarks compared to one-time alignment methods.
Jiabo Ye, Haiyang Xu, Haowei Liu +6 authors
mPLUG-Owl3, a versatile multi-modal large language model, introduces hyper attention blocks for improved integration of vision and language, achieving state-of-the-art performance on single-image, multi-image, and video tasks, particularly in scenarios with long visual sequences.
Zihan Qiu, Zeyu Huang, Shuang Cheng +4 authors
Layerwise Recurrent Router for Mixture-of-Experts (RMoE) improves MoE models by leveraging GRU-based cross-layer dependencies, enhancing expert selection and diversity without significant computational cost.
Yuxiang Zheng, Shichao Sun, Lin Qiu +13 authors
OpenResearcher is an AI-driven platform using RAG and LLMs to assist researchers by answering questions and retrieving domain-specific information efficiently and accurately.
Haider Al-Tahan, Quentin Garrido, Randall Balestriero +3 authors
UniBench provides a unified evaluation framework for vision-language models across various benchmarks, revealing that scaling does not always improve reasoning capabilities and simpler networks can outperform large models on basic tasks.
Robert Geirhos, Priyank Jaini, Austin Stone +5 authors
A visual memory system that combines deep neural networks with databases to offer flexible data addition, removal, and interpretable decision-making, enhancing knowledge representation in visual models.
Ruonan Yu, Songhua Liu, Zigeng Chen +2 authors
A novel label-lightening framework, HeLlO, reduces storage requirements for dataset distillation by generating labels online, using prior knowledge from foundation models and low-rank matrices.
Sean J. Liu, Nupur Kumari, Ariel Shamir +1 authors
A framework using Generative Photomontage allows users to composite images from different parts of generated images, enhancing flexibility and quality in text-to-image models with new blending techniques.
Jiasong Feng, Ao Ma, Jing Wang +4 authors
FancyVideo introduces the CTGM to improve temporal consistency in text-to-video synthesis by incorporating TII, TAR, and TFB for frame-specific text control.
Lukas Meyer, Andreas Gilson, Ute Schmidt +1 authors
FruitNeRF uses semantic neural radiance fields to count fruits from unordered images, achieving precise 3D fruit counting without type dependency.
Jiri Hron, Laura Culp, Gamaleldin Elsayed +28 authors
Larger language models trained on fixed datasets exhibit fewer hallucinations, but detecting these hallucinations becomes more difficult as model size increases, despite improvements in detector size.
Kaiser Sun, Mark Dredze
Investigating pre-train-then-fine-tune paradigms reveals that additional pre-training enhances model capabilities and fine-tuning amplifies performance on incapable tasks, but may cause forgetting of prior knowledge and increased sensitivity to evaluation prompts.
Xiao Liu, Tianjie Zhang, Yu Gu +27 authors
VisualAgentBench (VAB) is a benchmark designed to evaluate Large Multimodal Models (LMMs) across various real-world scenarios, including Embodied, Graphical User Interface, and Visual Design tasks, to assess their understanding and interaction capabilities.
Jiarui Lu, Thomas Holleis, Yizhe Zhang +9 authors
ToolSandbox evaluates large language models with tool-use capabilities through stateful execution, implicit state dependencies, and conversational simulation, revealing performance gaps and challenges with complex tasks.
Fanqi Wan, Longguang Zhong, Ziyi Yang +2 authors
FuseChat integrates diverse chat LLMs using lightweight fine-tuning and parameter-space merging, resulting in a highly capable model competitive with larger LLMs.
Yu Liu, Baoxiong Jia, Yixin Chen +1 authors
SlotLifter, a novel object-centric radiance model, achieves state-of-the-art performance in scene decomposition and novel-view synthesis by integrating object-centric learning with image-based rendering techniques.
Sarah Barrington, Matyas Bohacek, Hany Farid
We describe a large-scale dataset--{\em DeepSpeak}--of real and deepfake footage of people talking and gesturing in front of their webcams. The real videos in this first version of the dataset consist of 9 hours of footage from 220 diverse individuals. Constituting more than 25 hours of footage, the fake videos consist of a range of different state-of-the-art face-swap and lip-sync deepfakes with natural and AI-generated voices. We expect to release future versions of this dataset with different and updated deepfake technologies. This dataset is made freely available for research and non-commercial uses; requests for commercial use will be considered.
Bo-Wen Zhang, Liangdong Wang, Jijie Li +6 authors
The Aquila2 series of bilingual models, using the HeuriMentor framework, achieve competitive performance on English and Chinese benchmarks with efficient training and data management.
Bo-Wen Zhang, Yan Yan, Lin Li +1 authors
InfinityMATH is a scalable dataset for programmatic mathematical reasoning that decouples numbers from problems, enhancing model performance and robustness across benchmarks.
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