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。
An Yang, Anfeng Li, Baosong Yang +57 authors
Qwen3, a unified series of large language models, integrates thinking and non-thinking modes, reduces computational resources, and achieves state-of-the-art performance across various tasks and languages.
50 篇论文 · 按点赞排序
Chaorui Deng, Deyao Zhu, Kunchang Li +9 authors
BAGEL, an open-source foundational model trained on diverse multimodal data, significantly outperforms existing models in both generation and understanding tasks.
NovelSeek Team, Bo Zhang, Shiyang Feng +22 authors
Artificial Intelligence (AI) is accelerating the transformation of scientific research paradigms, not only enhancing research efficiency but also driving innovation. We introduce NovelSeek, a unified closed-loop multi-agent framework to conduct Autonomous Scientific Research (ASR) across various scientific research fields, enabling researchers to tackle complicated problems in these fields with unprecedented speed and precision. NovelSeek highlights three key advantages: 1) Scalability: NovelSeek has demonstrated its versatility across 12 scientific research tasks, capable of generating innovative ideas to enhance the performance of baseline code. 2) Interactivity: NovelSeek provides an interface for human expert feedback and multi-agent interaction in automated end-to-end processes, allowing for the seamless integration of domain expert knowledge. 3) Efficiency: NovelSeek has achieved promising performance gains in several scientific fields with significantly less time cost compared to human efforts. For instance, in reaction yield prediction, it increased from 27.6% to 35.4% in just 12 hours; in enhancer activity prediction, accuracy rose from 0.52 to 0.79 with only 4 hours of processing; and in 2D semantic segmentation, precision advanced from 78.8% to 81.0% in a mere 30 hours.
Kaitao Song, Xiaohua Wang, Xu Tan +14 authors
A novel Chain-of-Model framework introduces hierarchical hidden state chains in Transformers to improve scaling efficiency and inference flexibility for language models.
Hyungjoo Chae, Sunghwan Kim, Junhee Cho +18 authors
The paper introduces Web-Shepherd, a process reward model for web navigation, which improves accuracy and cost-effectiveness in step-level trajectory assessment compared to existing multimodal large language models.
Ling Yang, Ye Tian, Bowen Li +4 authors
MMaDA, a multimodal diffusion foundation model, achieves superior performance through a unified architecture, mixed long chain-of-thought fine-tuning, and a unified policy-gradient-based RL algorithm.
Jiajie Zhang, Nianyi Lin, Lei Hou +2 authors
AdaptThink, an RL algorithm, enhances reasoning models' efficiency and performance by adaptively selecting the optimal thinking mode based on problem difficulty.
Mengzhao Chen, Chaoyi Zhang, Jing Liu +8 authors
A unified scaling law for quantization-aware training (QAT) identifies key factors affecting quantization error, leading to improvements through mixed-precision quantization.
Jintao Zhang, Jia Wei, Pengle Zhang +6 authors
Efficiency enhancements for attention mechanisms, including leveraging FP4 Tensor Cores and developing an 8-bit attention method, improve inference and training performance.
Tingchen Fu, Jiawei Gu, Yafu Li +2 authors
An empirical analysis of MathIF identifies a tension between enhancing reasoning capacity and maintaining instruction adherence in large language models.
Yue Liu, Shengfang Zhai, Mingzhe Du +9 authors
GuardReasoner-VL enhances VLM safety through a reasoning-based guard model trained with SFT and online RL, using diverse datasets and a length-aware safety reward.
Guanting Dong, Yifei Chen, Xiaoxi Li +7 authors
Tool-Star, an RL-based framework, enables LLMs to autonomously use multiple tools for stepwise reasoning, leveraging data synthesis and hierarchical reward design.
Chenwei Lou, Zewei Sun, Xinnian Liang +6 authors
AdaCoT, an adaptive reasoning framework using reinforcement learning, reduces unnecessary Chain-of-Thought generation by LLMs, cutting computational costs without sacrificing performance on complex tasks.
Siyue Zhang, Yilun Zhao, Liyuan Geng +3 authors
Diffusion language models outperform large language model embeddings in text retrieval tasks due to their bidirectional architecture.
Yi Xu, Chengzu Li, Han Zhou +4 authors
Visual Planning, using sequences of images for planning, outperforms text-only reasoning methods in visual navigation tasks.
Alex Su, Haozhe Wang, Weimin Ren +2 authors
Introducing pixel-space reasoning in Vision-Language Models (VLMs) through visual operations like zoom-in and select-frame enhances their performance on visual tasks.
Sule Bai, Mingxing Li, Yong Liu +5 authors
UniVG-R1, a reasoning-guided multimodal large language model, enhances visual grounding by leveraging reinforcement learning and a difficulty-aware strategy, achieving state-of-the-art results and strong generalizability.
Gongfan Fang, Xinyin Ma, Xinchao Wang
Thinkless enables LLMs to adaptively choose between short and long reasoning by using control tokens, reducing computational inefficiencies on benchmarks.
Jeffrey Willette, Heejun Lee, Sung Ju Hwang
A novel procedure corrects distributional shifts in sparse attention, enhancing performance and reducing computational cost compared to full quadratic attention.
Tianbao Xie, Jiaqi Deng, Xiaochuan Li +12 authors
A new benchmark OSWorld-G and dataset Jedi enhance GUI grounding by capturing complex interactions, leading to improved performance in computer use agents.
Yongliang Wu, Zonghui Li, Xinting Hu +7 authors
KRIS-Bench assesses generative models' knowledge-based reasoning in image editing through a taxonomy of editing tasks and a Knowledge Plausibility metric.
Yanheng He, Jiahe Jin, Pengfei Liu
PC Agent-E framework improves data efficiency and achieves superior performance on human-like computer use tasks through enhanced trajectory synthesis and training.
Benjamin Schneider, Dongfu Jiang, Chao Du +2 authors
QuickVideo accelerates long-video understanding by combining a parallelized video decoder, memory-efficient prefilling, and overlapping video decoding with inference, enabling real-time performance.
Zhaowei Wang, Wenhao Yu, Xiyu Ren +9 authors
MMLongBench is a benchmark that evaluates long-context vision-language models across various tasks, image types, and input lengths, revealing that single-task performance is insufficient for gauging overall capability.
Ben Cohen, Emaad Khwaja, Youssef Doubli +14 authors
Toto is a time series forecasting foundation model using a decoder-only architecture, demonstrating state-of-the-art performance on large-scale benchmarks with multivariate observability data.
Yunshui Li, Yiyuan Ma, Shen Yan +21 authors
Model merging during pre-training enhances large language models, improving performance and reducing costs across various architectures and learning rates.
Jiaxin Guo, Zewen Chi, Li Dong +4 authors
RRMs, employing chain-of-thought reasoning and reinforcement learning, enhance reward model performance by adaptively utilizing test-time compute.
Peiyuan Zhang, Haofeng Huang, Yongqi Chen +5 authors
Trainable sparse attention (VSA) reduces computational cost in video diffusion transformers with minimal impact on performance, enabling efficient scaling of the models.
Yang Chen, Zhuolin Yang, Zihan Liu +5 authors
Large-scale reinforcement learning enhances reasoning capabilities in small and mid-sized models more effectively than distillation, achieving superior results in both math and code benchmarks.
Emile van Krieken, Pasquale Minervini, Edoardo Ponti +1 authors
Neurosymbolic diffusion models address limitations of standard neurosymbolic predictors by modeling dependencies between symbols using discrete diffusion, leading to improved accuracy and calibration.
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