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。
Xuekai Zhu, Daixuan Cheng, Dinghuai Zhang +20 authors
FlowRL enhances LLM reinforcement learning by matching the full reward distribution through flow balancing, improving diversity and performance over reward-maximizing methods.
50 篇论文 · 按点赞排序
Liangcai Su, Zhen Zhang, Guangyu Li +19 authors
AgentFounder, a deep research agent model incorporating Agentic Continual Pre-training, achieves state-of-the-art performance in agentic tasks while maintaining strong tool-use ability.
Zhaoyang Liu, JingJing Xie, Zichen Ding +18 authors
ScaleCUA, a large-scale dataset and model for computer use agents, achieves state-of-the-art performance across multiple platforms and tasks by leveraging data-driven scaling.
Zijian Li, Xin Guan, Bo Zhang +9 authors
WebWeaver, a dual-agent framework, addresses open-ended deep research challenges by integrating adaptive planning and focused synthesis to produce high-quality, reliable reports.
Yang Zhou, Yifan Wang, Jianjun Zhou +16 authors
OmniWorld, a large-scale, multi-domain, multi-modal dataset, addresses the limitations of existing 4D world modeling datasets and benchmarks, enabling significant performance improvements in 4D reconstruction and video generation.
Kuan Li, Zhongwang Zhang, Huifeng Yin +14 authors
WebSailor, a post-training methodology, enhances open-source models with systematic uncertainty reduction, matching proprietary agents' performance in complex information-seeking tasks.
Hasan Abed Al Kader Hammoud, Mohammad Zbeeb, Bernard Ghanem
Hala, a family of Arabic-centric instruction and translation models, achieves state-of-the-art results using a translate-and-tune pipeline, slerp merging, and fine-tuning on high-quality bilingual supervision.
Xixi Wu, Kuan Li, Yida Zhao +11 authors
ReSum, a novel paradigm with periodic context summarization, enhances web agents' performance on knowledge-intensive tasks by overcoming context window limitations, achieving significant improvements over ReAct.
Runnan Fang, Shihao Cai, Baixuan Li +13 authors
A scalable framework and two-phase fine-tuning strategy enhance function-calling capabilities of agents in diverse environments, improving performance on agentic benchmarks.
Zile Qiao, Guoxin Chen, Xuanzhong Chen +13 authors
WebResearcher, a deep-research framework, enhances AI agents' knowledge synthesis by reformulating research as a Markov Decision Process and using a scalable data synthesis engine, achieving superior performance across benchmarks.
Haoran Zhang, Yafu Li, Xuyang Hu +4 authors
Align3, a lightweight method using Test-Time Deliberation, enhances specification alignment in large language models across diverse scenarios with minimal overhead.
Zhengxi Lu, Jiabo Ye, Fei Tang +8 authors
Semi-online Reinforcement Learning addresses the limitations of offline and online RL by simulating online RL on offline trajectories, achieving state-of-the-art performance in dynamic benchmarks.
Weijie Yin, Yongjie Ye, Fangxun Shu +11 authors
SAIL-VL2, a vision-language foundation model, achieves state-of-the-art performance across diverse benchmarks through data curation, progressive training, and sparse MoE architecture.
Biwen Lei, Yang Li, Xinhai Liu +97 authors
Hunyuan3D Studio automates 3D asset creation using AI, integrating neural modules to transform concept images or text into high-quality, game-ready 3D models with optimized geometry and PBR textures.
Jiasen Lu, Liangchen Song, Mingze Xu +5 authors
AToken, a unified visual tokenizer, achieves high-fidelity reconstruction and semantic understanding across images, videos, and 3D assets using a 4D transformer architecture with adversarial-free training.
Akshit Sinha, Arvindh Arun, Shashwat Goel +2 authors
Scaling large language models improves their ability to execute longer tasks by isolating execution capability and mitigating self-conditioning effects, despite diminishing single-step accuracy.
Zhongwen Xu, Zihan Ding
Single-stream Policy Optimization (SPO) improves policy-gradient training for Large Language Models by eliminating group-based issues and providing a stable, low-variance learning signal, leading to better performance and efficiency.
Weichu Liu, Jing Xiong, Yuxuan Hu +11 authors
LongEmotion benchmark enhances large language models' emotional intelligence in long-context scenarios using Retrieval-Augmented Generation and Collaborative Emotional Modeling.
Yujun Zhou, Zhenwen Liang, Haolin Liu +7 authors
EVOL-RL, a label-free reinforcement learning method, enhances large language models by balancing stability and variation, preventing entropy collapse and improving generalization.
Chenxi Song, Yanming Yang, Tong Zhao +2 authors
WorldForge, a training-free framework, enhances video diffusion models with precise motion control and photorealistic content generation through recursive refinement, flow-gated latent fusion, and dual-path self-corrective guidance.
Liang Hu, Jianpeng Jiao, Jiashuo Liu +20 authors
FinSearchComp is an open-source benchmark for evaluating financial search and reasoning capabilities of end-to-end agents, featuring realistic tasks and professional annotations.
Weipeng Zhong, Peizhou Cao, Yichen Jin +9 authors
InternScenes is a large-scale, diverse, and realistic indoor scene dataset that addresses limitations in existing datasets, enabling better scene layout generation and point-goal navigation.
Tao Han, Wanghan Xu, Junchao Gong +4 authors
InfGen, a one-step generator replacing the VAE decoder, enables arbitrary high-resolution image generation from a fixed-size latent, significantly reducing computational complexity and generation time.
Xiaoyu Yue, Zidong Wang, Yuqing Wang +5 authors
Self-guided Training for AutoRegressive models (ST-AR) enhances image understanding and generation quality in autoregressive models by addressing key visual semantics challenges through self-supervised objectives.
Wenyan Li, Raphael Tang, Chengzu Li +3 authors
Two approaches are introduced to analyze and quantify information loss in vision-language models during the projection of visual inputs into the language model's embedding space, revealing significant distortions and their impact on model performance.
Xu Zheng, Chenfei Liao, Ziqiao Weng +12 authors
Recent advancements in omnidirectional vision, driven by industrial and academic interest, have led to breakthroughs in generation, perception, and understanding, with the proposal of a new system architecture called PANORAMA.
Xinhao Yan, Jiachen Xu, Yang Li +8 authors
X-Part is a generative model that decomposes 3D objects into semantically meaningful parts with high fidelity, using bounding boxes and point-wise semantic features, and supports interactive editing.
Nenad Tomasev, Matija Franklin, Joel Z. Leibo +4 authors
The sandbox economy framework analyzes the emerging AI agent economy, focusing on its origins and permeability, and discusses design choices for safe and steerable AI markets.
Duolin Sun, Dan Yang, Yue Shen +7 authors
HANRAG, a heuristic-based framework, improves question-answering systems by efficiently handling multi-hop queries and reducing noise through query decomposition and filtering.
Alejandro Hernández-Cano, Alexander Hägele, Allen Hao Huang +98 authors
Apertus is a suite of open large language models that ensure data compliance and multilingual representation through ethical data sourcing, the Goldfish objective, and comprehensive artifact release.
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