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。
50 篇论文 · 按点赞排序
Quang-Binh Nguyen, Minh Luu, Quang Nguyen +2 authors
CSD-VAR, a Visual Autoregressive Modeling approach, enhances content-style decomposition by introducing scale-aware optimization, SVD-based rectification, and augmented K-V memory, outperforming diffusion models in content preservation and stylization.
Hao Sun, Mihaela van der Schaar
A review of recent advances in aligning Large Language Models using inverse reinforcement learning, emphasizing the construction of neural reward models and addressing challenges in training and evaluation.
Xingyu Wu, Yuchen Yan, Shangke Lyu +7 authors
Length-Adaptive Policy Optimization (LAPO) reduces token usage and improves accuracy by enabling models to internally manage reasoning depth through reinforcement learning.
Shang Liu, Chenjie Cao, Chaohui Yu +3 authors
EarthCrafter, a framework using sparse-decoupled latent diffusion, efficiently generates large-scale 3D Earth models by separating structural and textural generation, leveraging the Aerial-Earth3D dataset.
Zuo-Liang Zhu, Jian Yang, Beibei Wang
A discretized signed distance field (SDF) is introduced to enhance Gaussian splatting for inverse rendering, improving relighting quality without additional memory or complex optimization.
Chuanhao Yan, Fengdi Che, Xuhan Huang +13 authors
Formal language-based reasoning and automatic verification improve the reliability and scalability of Large Language Models in generating verifiable code.
Junhao Shen, Haiteng Zhao, Yuzhe Gu +7 authors
SOPHIA, a semi-off-policy reinforcement learning approach, enhances large vision-language models with slow-thinking reasoning, improving performance on multimodal reasoning tasks.
Guowei Lan, Kaixian Qu, René Zurbrügg +4 authors
ExpTeach grounds vision-language models to physical robots through self-generated memory and retrieval-augmented generation, improving success rates and enabling intelligent object interactions.
Zhiwei Liu, Jielin Qiu, Shiyu Wang +9 authors
MCPEval is an open-source framework that automates task generation and evaluation for Large Language Models across various domains, improving the assessment process.
Aryo Pradipta Gema, Alexander Hägele, Runjin Chen +11 authors
Evaluation of Large Reasoning Models across different reasoning lengths reveals that increased test-time compute can lead to performance degradation and amplify problematic reasoning patterns.
Yuanhan Zhang, Yunice Chew, Yuhao Dong +3 authors
Video-TT assesses video LLMs' correctness and robustness in interpreting real-world videos through open-ended and adversarial questions.
Jiakang Wang, Runze Liu, Fuzheng Zhang +2 authors
Archer, an entropy-aware RLVR approach with dual-token constraints and synchronous updates, enhances LLM reasoning abilities by differentiating between knowledge and reasoning tokens, achieving state-of-the-art performance on mathematical reasoning and code generation benchmarks.
Zhekai Chen, Ruihang Chu, Yukang Chen +4 authors
TTS-VAR is a test-time scaling framework for visual auto-regressive models that improves generation quality by dynamically adjusting batch size and using clustering and resampling techniques.
Rithesh Murthy, Ming Zhu, Liangwei Yang +6 authors
Promptomatix automates prompt optimization for Large Language Models, improving performance and efficiency across various tasks.
Nuri Ryu, Jiyun Won, Jooeun Son +3 authors
Elevate3D enhances low-quality 3D assets by iteratively refining both texture and geometry using specialized methods and monocular geometry predictors, achieving state-of-the-art quality.
Jing Gu, Xian Liu, Yu Zeng +8 authors
PhyWorldBench evaluates video generation models' adherence to physical laws through a comprehensive benchmark, including an "Anti-Physics" category, and provides recommendations for improving physical realism.
Gen Luo, Wenhan Dou, Wenhao Li +9 authors
Mono-InternVL, an advanced monolithic Multimodal Large Language Model, integrates visual experts and employs Endogenous Visual Pre-training to enhance visual learning and reduce computational costs.
Dong Zhuo, Wenzhao Zheng, Jiahe Guo +3 authors
A streaming 4D visual geometry transformer uses causal attention and knowledge distillation to achieve real-time 4D reconstruction with high spatial consistency and competitive performance.
Shangke Lyu, Linjuan Wu, Yuchen Yan +7 authors
Hierarchical Budget Policy Optimization (HBPO) is a reinforcement learning framework that optimizes reasoning depth for large models, improving efficiency and accuracy by adapting to problem complexity.
Qiaoyu Tang, Hao Xiang, Le Yu +6 authors
RefCritic, a reinforcement learning-based critic module with dual rule-based rewards, enhances model critique and refinement capabilities, demonstrating superior performance across multiple benchmarks compared to supervised fine-tuning methods.
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