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
Ivan Rodkin, Daniil Orel, Konstantin Smirnov +9 authors
Models trained on random Boolean functions in a cellular automata framework show that increasing depth, recurrence, memory, and test-time compute scaling enhances multi-step reasoning capabilities.
Pengxiang Li, Yefan Zhou, Dilxat Muhtar +6 authors
Prophet, a training-free fast decoding paradigm for diffusion language models, reduces inference time by leveraging early answer convergence without sacrificing quality.
Kaiyue Sun, Rongyao Fang, Chengqi Duan +2 authors
T2I-ReasonBench evaluates the reasoning capabilities of text-to-image models across four dimensions using a two-stage protocol, analyzing their performance comprehensively.
Sixun Dong, Juhua Hu, Mian Zhang +3 authors
A multimodal method leverages both vision and text tokens to optimize vision token selection, improving inference efficiency in vision-language models.
Yang Zhou, Sunzhu Li, Shunyu Liu +10 authors
RuscaRL, a novel instructional scaffolding framework, enhances LLM reasoning by using rubrics for exploration and verifiable rewards, significantly improving performance on reasoning tasks.
Frano Rajič, Haofei Xu, Marko Mihajlovic +7 authors
A multi-view 3D point tracker using a feed-forward model with transformers and k-nearest-neighbors achieves robust tracking with fewer cameras and less optimization compared to existing methods.
Zayd M. K. Zuhri, Erland Hilman Fuadi, Alham Fikri Aji
Token Order Prediction (TOP) improves language model training by ordering upcoming tokens, outperforming both Next-Token Prediction (NTP) and Multi-Token Prediction (MTP) across benchmarks.
Xin Gao, Li Hu, Siqi Hu +20 authors
Wan-S2V, an audio-driven model built on Wan, enhances expressiveness and fidelity in cinematic character animation compared to existing methods.
Simin Ma, Shujian Liu, Jun Tan +7 authors
Task Centric Instruction Augmentation (TCIA) enhances large language models' performance on specific tasks while maintaining general instruction-following ability.
Yuxin Guo, Teng Wang, Yuying Ge +4 authors
AudioStory integrates large language models with text-to-audio systems to generate coherent, long-form audio narratives through a unified framework with decoupled bridging mechanisms and end-to-end training.
Jingqiao Xiu, Fangzhou Hong, Yicong Li +5 authors
EgoTwin, a diffusion transformer framework, addresses viewpoint alignment and causal interplay in joint egocentric video and human motion generation using a head-centric motion representation and cybernetics-inspired interaction mechanism.
Wei Xiong, Wenting Zhao, Weizhe Yuan +4 authors
A generative judge model, StepWiser, uses reinforcement learning to provide step-by-step reasoning feedback, improving both training and inference performance of policy models.
Haonan Qiu, Ning Yu, Ziqi Huang +2 authors
CineScale is a novel inference paradigm that enables high-resolution visual generation for both images and videos without extensive fine-tuning, addressing issues of repetitive patterns and high-frequency information accumulation.
Jeonghwan Kim, Yushi Lan, Armando Fortes +2 authors
A framework for efficient artistic mesh generation reduces redundancy by separating vertex and face generation, using an autoregressive model for vertices and a bidirectional transformer for faces, and includes a fidelity enhancer and post-processing to improve quality and speed.
Zhilin Zhang, Xiang Zhang, Jiaqi Wei +2 authors
PosterGen, a multi-agent framework using large language models, automates paper-to-poster generation with high design quality and minimal manual refinement.
Qianyu He, Siyu Yuan, Xuefeng Li +2 authors
ThinkDial is an open-source framework that implements controllable reasoning in large language models through discrete operational modes, achieving performance while reducing computational effort.
Fan Nie, Ken Ziyu Liu, Zihao Wang +11 authors
UQ is a benchmark for evaluating AI models on unsolved questions, combining difficulty and realism to assess capabilities like reasoning, factuality, and browsing.
Minghao Li, Ying Zeng, Zhihao Cheng +2 authors
ReportBench evaluates the content quality of research reports generated by large language models, focusing on cited literature quality and statement faithfulness, demonstrating that commercial Deep Research agents produce more comprehensive and reliable reports than standalone LLMs.
Harethah Abu Shairah, Hasan Abed Al Kader Hammoud, George Turkiyyah +1 authors
Rank-One Safety Injection (ROSI) enhances Large Language Model safety by amplifying refusal-mediating subspace activations without fine-tuning.
Qiaoyu Zheng, Yuze Sun, Chaoyi Wu +7 authors
Deep-DxSearch, an agentic RAG system trained with reinforcement learning, enhances medical diagnosis accuracy by integrating a large-scale retrieval corpus and tailored rewards.
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