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。
Guowei Xu, Peng Jin, Ziang Wu +5 authors
LLaVA-CoT is a vision-language model that achieves improved reasoning performance through structured multistage processing and test-time scaling, outperforming larger models with a smaller training dataset.
50 篇论文 · 按点赞排序
Weiyun Wang, Zhe Chen, Wenhai Wang +8 authors
A preference optimization process enhances multimodal reasoning, particularly Chain-of-Thought performance, in large language models through automated data construction and mixed preference optimization, demonstrating significant improvements on benchmarks.
Taiming Lu, Tianmin Shu, Alan Yuille +2 authors
Generative World Explorer (Genex) enables agents to mentally explore 3D environments using imagined observations, updating their beliefs without physical exploration to improve decision-making.
Yu Zhao, Huifeng Yin, Bo Zeng +6 authors
Marco-o1 extends reasoning models to open-ended domains without clear standards using Chain-of-Thought fine-tuning, Monte Carlo Tree Search, reflection mechanisms, and innovative reasoning strategies.
Maurice Weber, Daniel Fu, Quentin Anthony +16 authors
The RedPajama datasets are introduced to address core challenges for open-source language models by providing transparent data curation, large volumes of high-quality text, and quality signals for web data analysis.
Jintao Zhang, Haofeng Huang, Pengle Zhang +3 authors
SageAttention2 improves attention computation speed with 4-bit matrix multiplication and adaptive quantization, achieving significant OPS increases without compromising accuracy.
Xin Dong, Yonggan Fu, Shizhe Diao +10 authors
Hymba, a family of small language models with a hybrid-head architecture combining transformer attention and state space models, achieves state-of-the-art performance with improved efficiency and reduced cache size.
Enrico Fini, Mustafa Shukor, Xiujun Li +13 authors
AIMV2, a multimodal vision encoder paired with an autoregressive decoder, achieves superior performance in vision benchmarks and multimodal image understanding tasks compared to contrastive models.
Xudong Lu, Yinghao Chen, Cheng Chen +19 authors
BlueLM-V-3B optimizes the deployment of multimodal large language models on mobile platforms through redesigned dynamic resolution and hardware-aware system optimizations, achieving high performance with a small model size and fast generation speed.
Akari Asai, Jacqueline He, Rulin Shao +22 authors
OpenScholar, a specialized language model, enhances scientific query synthesis by retrieving relevant passages and providing citation-backed answers, outperforming existing models in correctness and citation accuracy.
Ziqi Huang, Fan Zhang, Xiaojie Xu +14 authors
VBench++ is a comprehensive benchmark suite for video generation that evaluates models across specific, hierarchical dimensions using fine-grained metrics and human preference annotations, offering insights into model strengths, weaknesses, and gaps compared to image generation models.
Siyuan Hu, Mingyu Ouyang, Difei Gao +1 authors
The recently released model, Claude 3.5 Computer Use, stands out as the first frontier AI model to offer computer use in public beta as a graphical user interface (GUI) agent. As an early beta, its capability in the real-world complex environment remains unknown. In this case study to explore Claude 3.5 Computer Use, we curate and organize a collection of carefully designed tasks spanning a variety of domains and software. Observations from these cases demonstrate Claude 3.5 Computer Use's unprecedented ability in end-to-end language to desktop actions. Along with this study, we provide an out-of-the-box agent framework for deploying API-based GUI automation models with easy implementation. Our case studies aim to showcase a groundwork of capabilities and limitations of Claude 3.5 Computer Use with detailed analyses and bring to the fore questions about planning, action, and critic, which must be considered for future improvement. We hope this preliminary exploration will inspire future research into the GUI agent community. All the test cases in the paper can be tried through the project: https://github.com/showlab/computer_use_ootb.
Zhennan Chen, Yajie Li, Haofan Wang +6 authors
RAG, a Regional-Aware text-to-image generation method, achieves fine-grained spatial control by decoupling generation into two sub-tasks, ensuring robust performance in real-world applications.
Xidong Feng, Ziyu Wan, Haotian Fu +6 authors
Natural Language Reinforcement Learning (NLRL) redefines traditional RL concepts in a language-based framework, leveraging large language models to achieve efficient and interpretable policy improvement.
Yuhao Dong, Zuyan Liu, Hai-Long Sun +4 authors
Insight-V enhances multi-modal large language models through scalable reasoning data generation and a multi-agent system, achieving performance improvements in visual and perceptual tasks.
Hmrishav Bandyopadhyay, Yi-Zhe Song
FlipSketch generates sketch animations by leveraging text-to-video diffusion models with fine-tuning, noise refinement, and dual-attention composition to create fluid and visually consistent animations from simple sketches and descriptions.
Yushi Lan, Shangchen Zhou, Zhaoyang Lyu +5 authors
A novel 3D generation framework using a VAE and cascaded latent diffusion model in a Point Cloud-structured Latent space achieves high-quality 3D generation with multi-modal inputs and disentangled shape-texture editing.
Guojun Lei, Chi Wang, Hong Li +3 authors
A unified controllable video generation method AnimateAnything utilizes a multi-scale fusion network and optical flows to handle various conditions, achieving high consistency and stability in generated videos.
Xinyan Guan, Yanjiang Liu, Xinyu Lu +8 authors
Verifier engineering uses automated verification feedback to enhance foundation models, proposing a critical step towards Artificial General Intelligence.
Chenxia Tang, Jianchun Liu, Hongli Xu +1 authors
A novel sampling method called top-$n\sigma$ for large language models improves reasoning task performance by filtering pre-softmax logits and maintaining consistent results across different temperatures.
Zihao Huang, Qiyang Min, Hongzhi Huang +4 authors
UltraMem introduces a large-scale, ultra-sparse memory layer to Transformer models, reducing inference latency and improving performance.
Omri Avrahami, Or Patashnik, Ohad Fried +4 authors
The work presents an automatic method to identify crucial layers in Diffusion Transformer models for consistent image editing and demonstrates its effectiveness through qualitative and quantitative evaluations.
Hongrui Jia, Chaoya Jiang, Haiyang Xu +6 authors
SymDPO optimizes preference for symbol demonstrations in large multimodal models, improving understanding and alignment in multimodal contexts.
Ziyang Luo, Haoning Wu, Dongxu Li +3 authors
VideoAutoArena and VideoAutoBench provide a scalable, cost-effective evaluation framework using automated user simulation and human-validated benchmarks to rigorously assess large multimodal models' video analysis capabilities.
Mathew Jacob, Erik Lindgren, Matei Zaharia +3 authors
Rerankers, typically cross-encoders, offer diminishing returns and degrade quality when scoring a large number of documents, challenging the assumption that they are consistently more effective.
Tianhe Ren, Yihao Chen, Qing Jiang +17 authors
DINO-X, a Transformer-based object-centric vision model, achieves top performance in open-world object detection through flexible prompts, a large-scale dataset, and integration of multiple perception tasks.
Cheng-Yen Yang, Hsiang-Wei Huang, Wenhao Chai +2 authors
SAMURAI enhances the Segment Anything Model 2 for visual object tracking by integrating motion cues and a memory selection mechanism, improving performance and achieving real-time zero-shot tracking without retraining.
Yu Gu, Boyuan Zheng, Boyu Gou +7 authors
WebDreamer enhances web agents' performance by using LLMs as world models to simulate action outcomes, leading to significant improvements over reactive baselines.
Haonan Wang, Qian Liu, Chao Du +4 authors
AnchorAttention improves the long-context performance and training efficiency of large language models by addressing numerical issues with BFloat16 and Rotary Positional Embedding.
Zili Wang, Robert Zhang, Kun Ding +3 authors
A speculative decoding algorithm is adapted for continuous-valued image generation to reduce inference time while preserving output quality.
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