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
Said Taghadouini, Adrien Cavaillès, Baptiste Aubertin
LightOnOCR-2-1B is a compact 1B-parameter vision-language model that performs end-to-end document image-to-text conversion with improved localization and robustness through specialized training techniques.
Jiayu Liu, Rui Wang, Qing Zong +7 authors
Large language models suffer from poor confidence calibration in retrieval-augmented generation due to noisy contexts, but a noise-aware calibration framework significantly improves calibration performance.
Xiangchi Yuan, Dachuan Shi, Chunhui Zhang +4 authors
Reinforced Agent Merging (RAM) addresses the limitations of traditional merging methods for reinforcement learning-trained agents by distinguishing shared and task-specific parameters to preserve critical behaviors during model integration.
Zecheng Tang, Baibei Ji, Ruoxi Sun +7 authors
A benchmark called MemoryRewardBench is introduced to systematically evaluate reward models' ability to assess long-term memory management in large language models across various context lengths and memory patterns.
Zhongxiang Sun, Yi Zhan, Chenglei Shen +4 authors
Personalized large language models can generate false information aligned with user history instead of factual truth, but a new method called FPPS helps maintain both factual accuracy and personalized responses while preserving existing personalization effects.
Linqing Zhong, Yi Liu, Yifei Wei +4 authors
Vision-Language-Action models are enhanced by incorporating action-space reasoning through a structured sequence of coarse action intents, improving manipulation task performance in both simulation and real-world environments.
Yawar Siddiqui, Duncan Frost, Samir Aroudj +9 authors
ShapeR generates high-fidelity 3D shapes from casual image sequences using visual-inertial SLAM, 3D detection, and vision-language models with rectified flow transformer conditioning.
Tanyu Chen, Tairan Chen, Kai Shen +4 authors
Chroma 1.0 enables real-time spoken dialogue with personalized voice cloning through discrete speech representations and interleaved text-audio token scheduling.
Letian Zhang, Sucheng Ren, Yanqing Liu +9 authors
An advanced vision encoder named OpenVision 3 learns a unified visual representation for both image understanding and generation by combining VAE-compressed image latents with ViT architecture and joint optimization of reconstruction and semantic signals.
Jing Zuo, Lingzhou Mu, Fan Jiang +3 authors
FantasyVLN presents a unified implicit reasoning framework for vision-and-language navigation that enhances reasoning capabilities without explicit token overhead, achieving real-time performance with improved accuracy.
Yang Yu, Peiyu Zang, Chi Hsu Tsai +11 authors
Large language models and agent-based systems are being leveraged to automate kernel generation and optimization, addressing the scalability challenges in hardware-specific code development through structured approaches and systematic benchmarking.
Dawei Li, Yuguang Yao, Zhen Tan +2 authors
ToolPRMBench is introduced as a large-scale benchmark for evaluating process reward models in tool-using agents, featuring step-level test cases and multi-LLM verification to ensure data quality.
Wenhan Liu, Xinyu Ma, Yutao Zhu +4 authors
A novel retriever training framework for agentic search that uses both local relevance and global answer correctness metrics with iterative optimization between the search agent and retriever.
Chuqiao Li, Xianghui Xie, Yong Cao +2 authors
A diffusion-based framework generates human motion from text prompts with fine-grained part-level control using a newly constructed dataset with atomic, temporally-aware annotations.
Gilat Toker, Nitay Calderon, Ohad Amosy +1 authors
A framework for generating structured counterfactual pairs using LLMs and SCMs enables improved evaluation and analysis of concept-based explanations in high-stakes domains.
Kanchana Ranasinghe, Honglu Zhou, Yu Fang +7 authors
A novel language-conditioned optical flow forecasting model combines Vision-Language Model and Diffusion architecture to predict future motion from noisy web-scale video data, demonstrating versatility in robotic manipulation and video generation tasks.
Yifan Wang, Shiyu Li, Peiming Li +3 authors
Render-of-Thought framework converts textual reasoning steps into images using vision-language models to improve reasoning traceability and efficiency while maintaining competitive performance.
Pedro Memoli Buffa, Luciano Del Corro
Output-entropy profiles computed from final-layer next-token probabilities serve as a scalable signal for monitoring LLM performance and prioritizing data acquisition under domain shifts.
Shiyu Liu, Yongjing Yin, Jianhao Yan +7 authors
Reinforcement learning framework for agentic search that improves reliability by teaching agents to recognize reasoning limits and respond appropriately when evidence is insufficient.
Sangbeom Lim, Seoung Wug Oh, Jiahui Huang +3 authors
VideoMaMa converts coarse masks to accurate alpha mattes using pretrained video diffusion models, enabling zero-shot generalization and scalable pseudo-labeling for video matting.
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