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。
Guanting Dong, Hangyu Mao, Kai Ma +11 authors
Agentic Reinforced Policy Optimization (ARPO) enhances multi-turn reasoning in large language models by balancing long-horizon capabilities and tool interactions, using entropy-based adaptive rollouts and advantage attribution.
50 篇论文 · 按点赞排序
HunyuanWorld Team, Zhenwei Wang, Yuhao Liu +52 authors
HunyuanWorld 1.0 generates immersive 3D scenes from text and images using a semantically layered 3D mesh representation with panoramic world proxies, achieving state-of-the-art performance in coherence, exploration, and interactivity.
Luoxin Chen, Jinming Gu, Liankai Huang +33 authors
Seed-Prover, a lemma-style reasoning model using Lean, achieves high performance in formal theorem proving and automated mathematical reasoning through iterative refinement and specialized geometry support.
Yilei Jiang, Yaozhi Zheng, Yuxuan Wan +4 authors
A modular multi-agent framework named ScreenCoder decomposes UI design-to-code translation into grounding, planning, and generation stages, achieving superior layout accuracy and code correctness through specialized agents and fine-tuned multimodal models.
Huan-ang Gao, Jiayi Geng, Wenyue Hua +24 authors
This survey reviews self-evolving agents, focusing on mechanisms for adaptation in components like models, memory, and architecture, and discusses challenges in safety, scalability, and co-evolutionary dynamics.
Jingwei Zuo, Maksim Velikanov, Ilyas Chahed +24 authors
Falcon-H1, a new series of large language models with a hybrid architecture combining Transformer-based attention and State Space Models, achieves state-of-the-art performance and efficiency across various tasks and sizes.
Rujun Han, Yanfei Chen, Zoey CuiZhu +15 authors
TTD-DR, a diffusion-based framework, generates high-quality research reports by iteratively refining a preliminary draft with external information and self-evolutionary algorithms, outperforming existing deep research agents.
Longwen Zhang, Qixuan Zhang, Haoran Jiang +4 authors
BANG is a generative approach using latent diffusion models and temporal attention to enable intuitive, part-level decomposition of 3D objects with precise control and multimodal interaction.
Yixin Song, Zhenliang Xue, Dongliang Wei +11 authors
SmallThinker, a family of LLMs designed for local devices, uses a deployment-aware architecture with sparse structures, pre-attention routing, and hybrid sparse attention to achieve high performance on consumer hardware.
Yuying Ge, Yixiao Ge, Chen Li +15 authors
ARC-Hunyuan-Video, a multimodal model, processes raw video inputs for structured comprehension, supporting captioning, summarization, question answering, grounding, and reasoning with high efficiency.
Ruifeng Yuan, Chenghao Xiao, Sicong Leng +9 authors
VL-Cogito, a multimodal reasoning model, uses a Progressive Curriculum Reinforcement Learning framework to improve performance across diverse tasks by dynamically adjusting difficulty and reasoning path length.
Miaosen Zhang, Ziqiang Xu, Jialiang Zhu +8 authors
The Phi-Ground model family achieves state-of-the-art performance in GUI grounding for multimodal reasoning models, improving accuracy across various benchmarks.
Zedong Wang, Siyuan Li, Dan Xu
Rep-MTL enhances multi-task learning by leveraging representation-level saliency to promote task complementarity and mitigate negative transfer.
Jiale Chen, Torsten Hoefler, Dan Alistarh
GPTQ quantization is mathematically equivalent to Babai's nearest plane algorithm, providing a geometric interpretation and error bounds for large language model quantization.
Zigang Geng, Yibing Wang, Yeyao Ma +10 authors
Reinforcement learning enhances discrete autoregressive modeling for image and language generation, achieving high-quality image generation and instruction-following capabilities using a unified framework.
Chao Yi, Dian Chen, Gaoyang Guo +50 authors
RecGPT integrates large language models into recommender systems to focus on user intent, improving content diversity and satisfaction while enhancing merchant and platform performance.
Yung-Sung Chuang, Yang Li, Dong Wang +13 authors
MetaCLIP 2, trained on worldwide web-scale image-text pairs, improves zero-shot classification and multilingual benchmarks without system-level confounding factors.
Yukang Cao, Jiahao Lu, Zhisheng Huang +8 authors
A survey organizes 4D spatial intelligence reconstruction methods into five levels, from low-level 3D attributes to incorporating physical laws, and discusses challenges and future directions.
Lakshya A Agrawal, Shangyin Tan, Dilara Soylu +14 authors
GEPA, a prompt optimizer using natural language reflection, outperforms RL methods like GRPO and MIPROv2 with fewer rollouts by learning high-level rules from trial and error.
Runjin Chen, Andy Arditi, Henry Sleight +2 authors
Persona vectors in large language models can monitor and control personality changes during training and deployment, enabling the identification and mitigation of undesirable traits.
Gyeongjin Kang, Seungtae Nam, Xiangyu Sun +3 authors
iLRM, an iterative Large 3D Reconstruction Model, improves scalability and efficiency in 3D reconstruction by decoupling scene representation, using a two-stage attention scheme, and injecting high-resolution information.
Yuzhong Zhao, Yue Liu, Junpeng Liu +9 authors
Geometric-Mean Policy Optimization (GMPO) stabilizes policy updates in large language models by maximizing the geometric mean of token-level rewards, outperforming Group Relative Policy Optimization (GRPO) on mathematical and multimodal reasoning benchmarks.
Xuehui Wang, Zhenyu Wu, JingJing Xie +25 authors
MMBench-GUI evaluates GUI automation agents across multiple platforms using a hierarchical benchmark and Efficiency-Quality Area metric, highlighting the importance of visual grounding, task planning, and efficiency.
Kele Shao, Keda Tao, Kejia Zhang +7 authors
A survey of token compression techniques for multimodal large language models, focusing on image, video, and audio data, categorizing methods by modality and mechanism.
Yumeng Wang, Zhiyuan Fan, Jiayu Liu +1 authors
A diversity-enhanced framework with multiple role perspectives improves accuracy and diversity in subjective reasoning tasks through unsupervised data construction and reinforcement learning with reward shaping.
Chengqian Ma, Wei Tao, Yiwen Guo
A benchmark dataset for Spoken Dialogue Models (SDMs) in English and Chinese is presented to evaluate their performance in understanding and emulating human conversations, addressing challenges like ambiguity and context-dependency.
Xiaoyu Chen, Hangxing Wei, Pushi Zhang +9 authors
The ViLLA framework enhances VLA models by incorporating latent actions, improving performance in both simulated and real-world robot manipulation tasks.
Xiaoya Li, Xiaofei Sun, Albert Wang +2 authors
CUDA-L1, an automated reinforcement learning framework, significantly improves CUDA optimization across various GPU architectures, achieving substantial speedups without human expertise.
Yuhan Wang, Siwei Yang, Bingchen Zhao +4 authors
A publicly available dataset, GPT-IMAGE-EDIT-1.5M, enhances instruction-guided image editing by fine-tuning open-source models to achieve competitive performance with proprietary models.
Dingkun Liu, Zhu Chen, Jingwei Luo +2 authors
MIRepNet, an EEG foundation model tailored for motor imagery, achieves state-of-the-art performance across multiple datasets using a hybrid pretraining strategy that combines self-supervised and supervised learning.
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