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。
Oriane Siméoni, Huy V. Vo, Maximilian Seitzer +23 authors
DINOv3, a self-supervised learning model, achieves superior performance across various vision tasks by scaling datasets and models, addressing dense feature degradation, and enhancing flexibility with post-hoc strategies.
50 篇论文 · 按点赞排序
Lei Bai, Zhongrui Cai, Maosong Cao +172 authors
Intern-S1, a multimodal Mixture-of-Experts model with extensive pre-training and reinforcement learning, achieves top-tier performance in general reasoning and outperforms closed-source models in scientific tasks.
Weizhen Li, Jianbo Lin, Zhuosong Jiang +27 authors
Chain-of-Agents (CoA) paradigm enables end-to-end complex problem-solving in LLMs through dynamic agent activation, improving performance via multi-agent distillation and agentic reinforcement learning.
Shiyin Lu, Yang Li, Yu Xia +39 authors
Ovis2.5, a native-resolution vision transformer with multimodal reasoning, achieves state-of-the-art performance on various benchmarks through advanced training techniques and efficient scaling methods.
Yuchen Fan, Kaiyan Zhang, Heng Zhou +15 authors
LLMs can serve as efficient simulators for RL tasks by leveraging internal knowledge, reducing reliance on external search engines and improving sim-to-real transfer.
Yichao Fu, Xuewei Wang, Yuandong Tian +1 authors
DeepConf enhances reasoning efficiency and performance by filtering low-quality reasoning traces using model-internal confidence signals, achieving high accuracy and reducing token generation.
Shuaijie She, Yu Bao, Yu Lu +7 authors
DuPO is a dual learning framework that generates annotation-free feedback using a generalized duality, enhancing performance across various tasks without relying on costly labels.
Yi-Fan Zhang, Xingyu Lu, Shukang Yin +17 authors
Thyme, a novel paradigm, enables MLLMs to autonomously perform image manipulations and computations, enhancing performance in perception and reasoning tasks through a two-stage training strategy and GRPO-ATS algorithm.
Juyuan Wang, Rongchen Zhao, Wei Wei +5 authors
ComoRAG, an iterative retrieval-based approach, enhances long-context narrative comprehension by dynamically updating memory and generating probing queries, outperforming traditional RAG methods.
Zhiyuan Zeng, Jiashuo Liu, Siyuan Chen +27 authors
FutureX is a dynamic, live benchmark for evaluating LLM agents in future prediction tasks, addressing challenges in real-time updates and data contamination.
Bingquan Dai, Li Ray Luo, Qihong Tang +9 authors
MeshCoder reconstructs complex 3D objects from point clouds into editable Blender Python scripts, enhancing shape-to-code reconstruction and 3D shape understanding through a multimodal large language model.
Jiabo Ye, Xi Zhang, Haiyang Xu +12 authors
GUI-Owl and Mobile-Agent-v3 are open-source GUI agent models and frameworks that achieve state-of-the-art performance across various benchmarks using innovations in environment infrastructure, agent capabilities, and scalable reinforcement learning.
Zhaoxi Chen, Tianqi Liu, Long Zhuo +6 authors
4DNeX generates high-quality dynamic 3D scene representations from a single image using a fine-tuned pretrained video diffusion model, outperforming existing methods in efficiency and generalizability.
Pratyush Maini, Vineeth Dorna, Parth Doshi +26 authors
BeyondWeb, a synthetic data generation framework, outperforms existing datasets and accelerates training for large language models by optimizing multiple factors.
Chin-Yang Lin, Cheng Sun, Fu-En Yang +3 authors
LongSplat improves novel view synthesis from long videos with irregular motion through joint optimization, robust pose estimation, and efficient anchor formation.
Ziyan Kuang, Feiyu Zhu, Maowei Jiang +8 authors
FinCDM, a cognitive diagnosis framework, evaluates financial LLMs at the knowledge-skill level using a comprehensive dataset, revealing hidden knowledge gaps and supporting more trustworthy model development.
Weigao Sun, Jiaxi Hu, Yucheng Zhou +12 authors
This survey examines innovative architectures for large language models to enhance efficiency, covering linear and sparse sequence modeling, efficient attention mechanisms, sparse mixture-of-experts, hybrid models, and diffusion LLMs.
NVIDIA, Aarti Basant, Abhijit Khairnar +208 authors
Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer model, enhances reasoning workload throughput and accuracy by replacing self-attention layers with Mamba-2 layers and using the Minitron strategy for compression.
Yuge Zhang, Nan Chen, Jiahang Xu +1 authors
POML addresses challenges in prompting Large Language Models by providing a structured, data-integrated, and format-sensitive markup language with templating and developer tools.
Yikai Wang, Zhouxia Wang, Zhonghua Wu +3 authors
A novel Next Visual Granularity (NVG) framework generates images by iteratively refining a sequence of visual granularities, outperforming existing methods in class-conditional image generation.
Chubin Chen, Jiashu Zhu, Xiaokun Feng +6 authors
S^2-Guidance, a novel method using stochastic block-dropping, improves sample quality and prompt adherence in diffusion models by refining suboptimal predictions, outperforming Classifier-free Guidance and other advanced strategies.
Ming Yin, Dinghan Shen, Silei Xu +11 authors
LiveMCP-101 benchmarks AI agents' ability to use multiple tools in real-world scenarios, revealing challenges in tool orchestration and inefficiencies in token usage.
Ziyang Luo, Zhiqi Shen, Wenzhuo Yang +7 authors
MCP-Universe is a comprehensive benchmark designed to evaluate large language models in realistic tasks through interaction with real-world MCP servers, addressing challenges like long-horizon reasoning and unfamiliar tool spaces.
Canyu Zhao, Xiaoman Li, Tianjian Feng +3 authors
Tinker is a framework for high-fidelity 3D editing using pretrained diffusion models, enabling multi-view consistency with minimal per-scene training.
Aditya Tomar, Coleman Hooper, Minjae Lee +7 authors
XQuant and XQuant-CL reduce memory consumption in LLM inference through low-bit quantization and cross-layer similarity exploitation, achieving significant memory savings with minimal accuracy loss.
Mikhail Seleznyov, Mikhail Chaichuk, Gleb Ershov +3 authors
A systematic evaluation of prompt robustness methods for Large Language Models across various models and tasks reveals insights into their effectiveness against format perturbations.
Yifu Zhang, Hao Yang, Yuqi Zhang +7 authors
Waver, a high-performance foundation model, generates high-quality videos and images using a Hybrid Stream DiT architecture, excelling in text-to-video, image-to-video, and text-to-image tasks.
Zhongang Cai, Yubo Wang, Qingping Sun +19 authors
Recent multi-modal models, including GPT-5, show significant progress in spatial intelligence but still lag behind human performance across various benchmarks.
Jiaqi Wei, Yuejin Yang, Xiang Zhang +19 authors
Agentic Science leverages large language models, multimodal systems, and integrated platforms to enable autonomous scientific discovery across various domains, encompassing hypothesis generation, experimental design, execution, analysis, and iterative refinement.
Xianglong He, Chunli Peng, Zexiang Liu +16 authors
Matrix-Game 2.0 generates real-time interactive videos using few-step auto-regressive diffusion, addressing the limitations of lengthy inference in existing models.
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