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。
Weihao Tan, Xiangyang Li, Yunhao Fang +11 authors
Lumine, a vision-language model-based agent, completes complex missions in real-time across different 3D open-world environments with human-like efficiency and zero-shot cross-game generalization.
50 篇论文 · 按点赞排序
Sen Xu, Yi Zhou, Wei Wang +7 authors
VibeThinker-1.5B, a 1.5B-parameter model using the Spectrum-to-Signal Principle, achieves superior reasoning capabilities compared to larger models at a significantly lower cost.
Aleksandr Razin, Danil Kazantsev, Ilya Makarov
LUA is a lightweight module that performs super-resolution directly in the latent space of diffusion models, improving efficiency without compromising image quality.
Jingyu Liu, Xin Dong, Zhifan Ye +6 authors
TiDAR combines diffusion and autoregressive models to achieve high throughput and quality in language generation.
Aarash Feizi, Shravan Nayak, Xiangru Jian +14 authors
GroundCUA, a large-scale desktop grounding dataset, enables the development of GroundNext models that achieve state-of-the-art performance in mapping instructions to UI elements with less training data.
Haotong Lin, Sili Chen, Junhao Liew +5 authors
Depth Anything 3 (DA3) uses a plain transformer for geometry prediction from visual inputs, achieving state-of-the-art results in camera pose estimation, any-view geometry, visual rendering, and monocular depth estimation.
Ding Chen, Simin Niu, Kehang Li +6 authors
HaluMem, a benchmark for evaluating memory hallucinations in AI systems, identifies and analyzes hallucinations across memory extraction, updating, and question answering stages using large-scale human-AI interaction datasets.
PAN Team, Jiannan Xiang, Yi Gu +31 authors
PAN, a general, interactable, and long-horizon world model, predicts future world states using a Generative Latent Prediction (GLP) architecture that combines autoregressive latent dynamics with a video diffusion decoder, enabling detailed, long-term, and coherent video simulation.
Guoxin Chen, Zile Qiao, Xuanzhong Chen +13 authors
IterResearch, an iterative deep-research paradigm, improves long-horizon reasoning by reformulating it as a Markov Decision Process with strategic workspace reconstruction and Efficiency-Aware Policy Optimization, achieving better performance and interaction scaling compared to existing agents.
Assaf Singer, Noam Rotstein, Amir Mann +2 authors
Time-to-Move (TTM) is a plug-and-play framework for motion- and appearance-controlled video generation using image-to-video (I2V) diffusion models, offering precise control over video content without requiring additional training.
Gleb V. Solovev, Alina B. Zhidkovskaya, Anastasia Orlova +18 authors
MADD, a multi-agent system integrating large language models and specialized models, streamlines hit identification in early drug discovery with superior performance and accessibility.
Zihao Yi, Qingxuan Jiang, Ruotian Ma +8 authors
LLMs struggle to authentically portray morally ambiguous or villainous characters due to safety alignment, as evidenced by the Moral RolePlay benchmark.
Tianzhu Ye, Li Dong, Zewen Chi +3 authors
Generative Adversarial Distillation (GAD) enhances black-box distillation by framing the student model as a generator and using a discriminator to provide adaptive feedback, surpassing traditional sequence-level knowledge distillation.
Speed Zhu, Jianwei Cai, Guang Chen +3 authors
The study presents a two-stage reinforcement learning approach for competitive-programming code generation, achieving state-of-the-art performance using Group Relative Policy Optimization and a hard-focus curriculum.
Rui Yang, Ziyu Zhu, Yanwei Li +9 authors
A framework called Visual Spatial Tuning (VST) enhances the spatial abilities of Vision-Language Models (VLMs) through progressive training with specialized datasets, achieving state-of-the-art results on spatial benchmarks.
Jack Hong, Chenxiao Zhao, ChengLin Zhu +3 authors
DeepEyesV2, an agentic multimodal model, uses a two-stage training pipeline to effectively integrate tool use, demonstrating robust performance across real-world reasoning tasks.
Soyeong Jeong, Aparna Elangovan, Emine Yilmaz +1 authors
A multi-agent framework enhances conversational quality by refining responses through agents responsible for factuality, personalization, and coherence, outperforming existing methods on challenging datasets.
Junghwan Lim, Sungmin Lee, Dongseok Kim +22 authors
Motif-2-12.7B combines architectural innovations and system optimizations to enhance efficiency and performance in large language models.
Zhengyang Liang, Daoan Zhang, Huichi Zhou +8 authors
UniVA is an open-source multi-agent framework that integrates video understanding, segmentation, editing, and generation into cohesive workflows using a Plan-and-Act architecture and hierarchical memory.
Hanqing Zhu, Zhenyu Zhang, Hanxian Huang +11 authors
Reinforcement Learning with Verifiable Rewards (RLVR) improves large language models by updating a limited set of parameters, which is explained by a Three-Gate Theory, revealing distinct optimization dynamics compared to supervised fine-tuning.
Stephen Chung, Wenyu Du
AI agents in the STATION environment achieve state-of-the-art performance across various benchmarks through autonomous scientific discovery and emergent behavior.
Seo Hyun Kim, Sunwoo Hong, Hojung Jung +2 authors
KL-Adaptive Stability Sampling (KLASS) accelerates diffusion-based generation by identifying stable predictions, achieving significant speedups and quality improvements across various domains.
Yu Feng, Nathaniel Weir, Kaj Bostrom +5 authors
VeriCoT, a neuro-symbolic method, formalizes and verifies logical arguments in Chain-of-Thought reasoning to improve the reliability and accuracy of LLMs.
Yingfeng Luo, Ziqiang Xu, Yuxuan Ouyang +9 authors
LMT, a suite of large-scale multilingual translation models, addresses challenges in multilingual machine translation through strategic downsampling and parallel multilingual prompting, achieving state-of-the-art performance across 60 languages.
Khalil Hennara, Ahmad Bastati, Muhammad Hreden +4 authors
A pipeline for processing the Common Crawl dataset to create a new Arabic multimodal dataset that preserves document structure and supports both text-only and multimodal pre-training.
Jiageng Mao, Sicheng He, Hao-Ning Wu +9 authors
PhysWorld integrates video generation and physical world modeling to enable accurate robotic manipulation from visual demonstrations without real robot data.
Nikolay Blagoev, Oğuzhan Ersoy, Lydia Yiyu Chen
The study identifies and defends against adversarial attacks in decentralized Group Relative Policy Optimization (GRPO) for Large Language Models (LLMs), demonstrating attack success rates of up to 100% and proposing effective defense mechanisms.
Eyal Gutflaish, Eliran Kachlon, Hezi Zisman +8 authors
A text-to-image model trained on long structured captions with DimFusion fusion mechanism and TaBR evaluation protocol achieves state-of-the-art prompt alignment and improved controllability.
Zhongyang Li, Ziyue Li, Tianyi Zhou
Aligning routing weights with task embeddings in Sparse Mixture-of-Experts (MoE) models improves generalization and reduces performance gaps in large language models.
Zefeng He, Xiaoye Qu, Yafu Li +3 authors
A novel video self-supervised reinforcement learning framework, VideoSSR, enhances MLLM performance across various video understanding tasks by leveraging intrinsic video information.
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