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。
Jingqi Tong, Yurong Mou, Hangcheng Li +11 authors
The "Thinking with Video" paradigm enhances multimodal reasoning by integrating video generation models, demonstrated through the Video Thinking Benchmark and improved performance on both vision and text tasks.
50 篇论文 · 按点赞排序
Jinjie Ni, Qian Liu, Longxu Dou +5 authors
Diffusion language models outperform autoregressive models in low-data settings due to any-order modeling, iterative bidirectional denoising, and Monte Carlo augmentation, and maintain advantages even at scale.
Nikita Kachaev, Mikhail Kolosov, Daniil Zelezetsky +2 authors
Systematic study reveals that naive action fine-tuning degrades visual representations in Vision-Language-Action models, but targeted strategies can mitigate this and improve generalization.
Kevin Qinghong Lin, Yuhao Zheng, Hangyu Ran +5 authors
VCode introduces a benchmark for generating SVG code from images to preserve symbolic meaning, highlighting gaps in visual-centric coding and proposing VCoder to improve performance.
Runqi Qiao, Qiuna Tan, Minghan Yang +10 authors
V-Thinker, a multimodal reasoning assistant using reinforcement learning, enhances image-interactive thinking by synthesizing datasets and aligning perception for improved performance in vision-centric tasks.
Jiawei Gu, Yunzhuo Hao, Huichen Will Wang +5 authors
ThinkMorph, a unified model fine-tuned on interleaved reasoning traces, enhances multimodal reasoning by generating coherent text-image steps, achieving significant performance gains and demonstrating emergent capabilities.
Ling-Team, Ang Li, Ben Liu +139 authors
Ling 2.0, a reasoning-oriented language model series, achieves high efficiency and accuracy through a Mixture-of-Experts paradigm, sparse activation, and innovative training techniques.
Zhaorun Chen, Zhuokai Zhao, Kai Zhang +15 authors
DreamGym is a unified framework that synthesizes diverse experiences for scalable online RL training, improving agent performance and reducing real-world interactions.
Mengzhao Chen, Meng Wu, Hui Jin +10 authors
A comprehensive comparison of low-precision floating-point and integer quantization in large language models reveals that fine-grained integer formats, especially MXINT8, offer superior accuracy and efficiency over floating-point formats in many cases.
Chenze Shao, Darren Li, Fandong Meng +1 authors
Continuous Autoregressive Language Models (CALM) improve language model efficiency by predicting continuous vectors instead of discrete tokens, reducing computational cost while maintaining performance.
Qiushi Sun, Mukai Li, Zhoumianze Liu +11 authors
A hybrid safety detection framework combining formal verification and VLM-based contextual assessment improves the detection of unsafe operations in mobile agents.
Kang Chen, Zhihao Liu, Tonghe Zhang +10 authors
The framework π<sub>RL</sub> uses reinforcement learning to train flow-based Vision-Language-Action models, addressing challenges with intractable action log-likelihoods and achieving significant performance improvements over supervised fine-tuning.
Yiyang Zhou, Haoqin Tu, Zijun Wang +11 authors
MIRA is a benchmark that evaluates models using intermediate visual images to enhance reasoning, showing significant performance improvements over text-only methods.
Guozhen Zhang, Zixiang Zhou, Teng Hu +6 authors
UniAVGen, a unified framework using dual Diffusion Transformers and Asymmetric Cross-Modal Interaction, enhances audio-video generation by ensuring synchronization and consistency with fewer training samples.
Travis Davies, Yiqi Huang, Alexi Gladstone +5 authors
EBT-Policy, an energy-based architecture, outperforms diffusion-based policies in robotic tasks by offering improved robustness, reduced computational cost, and emergent zero-shot recovery capabilities.
Gyeom Hwangbo, Hyungjoo Chae, Minseok Kang +3 authors
LEGO-Eval and LEGO-Bench improve the evaluation and generation of realistic 3D scenes by aligning detailed instructions with scene components, outperforming existing methods.
NVIDIA, Arslan Ali, Junjie Bai +86 authors
Cosmos-Predict2.5 and Cosmos-Transfer2.5 are advanced Physical AI models that unify text, image, and video generation, improve video quality and instruction alignment, and enable Sim2Real and Real2Real world translation with higher fidelity.
Fali Wang, Jihai Chen, Shuhua Yang +7 authors
Agent-REINFORCE optimizes multi-LLM collaboration graphs for test-time scaling, improving sample efficiency and search performance under accuracy and latency constraints.
Shusheng Yang, Jihan Yang, Pinzhi Huang +12 authors
Progress in multimodal intelligence requires a shift to supersensing, including semantic perception, event cognition, spatial cognition, and predictive modeling, demonstrated through VSI-SUPER benchmarks and a self-supervised predictive sensing approach.
Feng Han, Yibin Wang, Chenglin Li +8 authors
UniREditBench is a unified benchmark for reasoning-based image editing that addresses limitations in existing benchmarks by including multi-object interactions, game-world scenarios, and multimodal dual-reference evaluation.
Ropeway Liu, Hangjie Yuan, Bo Dong +6 authors
UniLumos, a unified relighting framework, enhances physical plausibility by integrating RGB-space geometry feedback into a flow matching backbone, achieving state-of-the-art results with improved consistency and speed.
Xinjian Zhao, Wei Pang, Zhongkai Xue +6 authors
Vision models outperform Graph Neural Networks on tasks requiring global structural understanding and scale-invariant reasoning, as demonstrated by the new GraphAbstract benchmark.
Joonghyuk Shin, Zhengqi Li, Richard Zhang +4 authors
MotionStream enables real-time video generation with sub-second latency and up to 29 FPS by distilling a text-to-video model with motion control into a causal student using Self Forcing with Distribution Matching Distillation and sliding-window causal attention with attention sinks.
NVIDIA, Amala Sanjay Deshmukh, Kateryna Chumachenko +123 authors
Nemotron Nano V2 VL, a hybrid Mamba-Transformer LLM, improves document and video understanding through enhanced architecture and token reduction techniques.
Yongyuan Liang, Wei Chow, Feng Li +7 authors
ROVER is a benchmark that evaluates reciprocal cross-modal reasoning in unified multimodal models, showing that cross-modal interactions significantly impact visual generation quality and that models struggle with symbolic reasoning tasks.
Penghui Qi, Zichen Liu, Xiangxin Zhou +4 authors
Using FP16 precision in reinforcement learning fine-tuning of large language models improves stability, convergence, and performance by addressing numerical mismatches.
Xinming Wang, Jian Xu, Bin Yu +9 authors
MR-ALIGN, a Meta-Reasoning informed alignment framework, enhances the factuality of large reasoning models by aligning their reasoning process, improving accuracy and reducing misleading reasoning.
Yuhong Liu, Beichen Zhang, Yuhang Zang +6 authors
Spatial-SSRL, a self-supervised reinforcement learning paradigm, enhances spatial understanding in Large Vision-Language Models using verifiable signals from RGB or RGB-D images without human annotation.
Kyungmin Lee, Sibeen Kim, Minho Park +4 authors
PHUMA, a large-scale and physically reliable humanoid locomotion dataset, improves motion imitation by addressing physical artifacts in human video data.
Chao Yan, Boyong Wu, Peng Yang +10 authors
Step-Audio-EditX, an open-source LLM-based audio model, excels in expressive and iterative audio editing and zero-shot TTS using large-margin synthetic data.
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