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。
GLM-4. 5 Team, Aohan Zeng, Xin Lv +168 authors
GLM-4.5, a Mixture-of-Experts large language model with 355B parameters, achieves strong performance across agentic, reasoning, and coding tasks using multi-stage training and reinforcement learning.
50 篇论文 · 按点赞排序
Runqi Qiao, Qiuna Tan, Peiqing Yang +11 authors
We-Math 2.0 enhances MLLMs' mathematical reasoning through a structured knowledge system, model-centric data space modeling, and reinforcement learning, demonstrating competitive performance on benchmarks.
NextStep Team, Chunrui Han, Guopeng Li +47 authors
NextStep-1, a 14B autoregressive model with a 157M flow matching head, achieves state-of-the-art performance in text-to-image generation and image editing by processing discrete text tokens and continuous image tokens.
Xinyu Geng, Peng Xia, Zhen Zhang +11 authors
WebWatcher, a multimodal agent with enhanced visual-language reasoning, outperforms existing agents in complex visual and textual information retrieval tasks using synthetic trajectories and reinforcement learning.
Wenhan Liu, Xinyu Ma, Weiwei Sun +4 authors
A reasoning-intensive reranker, ReasonRank, achieves state-of-the-art performance in passage ranking tasks by using synthesized training data and a two-stage post-training approach with reinforcement learning.
Ryan Wong, Jiawei Wang, Junjie Zhao +10 authors
WideSearch is a new benchmark evaluating the reliability of automated search agents in large-scale information collection tasks, revealing significant deficiencies in current systems.
Jinyuan Fang, Yanwen Peng, Xi Zhang +12 authors
A survey of self-evolving AI agents that adapt to dynamic environments through automatic enhancement based on interaction data and feedback.
Mo Yu, Tsz Ting Chung, Chulun Zhou +8 authors
A benchmark called PRELUDE evaluates long-context understanding by assessing the consistency of prequel stories with original books, revealing significant challenges for models compared to humans.
David Dinkevich, Matan Levy, Omri Avrahami +2 authors
Story2Board generates expressive storyboards from natural language using a consistency framework that enhances coherence and diversity without fine-tuning.
Lin Long, Yichen He, Wentao Ye +5 authors
M3-Agent, a multimodal agent with long-term memory, performs multi-turn reasoning and outperforms baselines on a new long-video question answering benchmark.
Seungyong Lee, Jeong-gi Kwak
Voost, a unified diffusion transformer framework, jointly learns virtual try-on and try-off, enhancing garment-body correspondence and achieving state-of-the-art results across benchmarks.
Lingen Li, Guangzhi Wang, Zhaoyang Zhang +6 authors
ToonComposer is a generative model that unifies inbetweening and colorization in cartoon production, using sparse sketches and a cartoon adaptation method to improve visual quality and efficiency.
Fangyuan Mao, Aiming Hao, Jintao Chen +7 authors
Omni-Effects is a unified framework that enables the generation of prompt-guided and spatially controllable composite visual effects using LoRA-based Mixture of Experts and Spatial-Aware Prompt with Independent-Information Flow.
Jiaxuan Gao, Wei Fu, Minyang Xie +5 authors
ASearcher is an open-source project that uses scalable asynchronous RL training to enhance search agents, achieving high performance on QA tasks with long-horizon search capabilities.
Zhongqi Yang, Wenhang Ge, Yuqi Li +12 authors
Matrix-3D generates wide-coverage 3D worlds from single images or text using panoramic video diffusion and reconstruction models.
Zihe Liu, Jiashun Liu, Yancheng He +12 authors
A systematic review of reinforcement learning techniques for large language model reasoning reveals clear guidelines and demonstrates that a minimalist combination of techniques can improve performance over existing strategies.
Nikita Dragunov, Temurbek Rahmatullaev, Elizaveta Goncharova +2 authors
SONAR-LLM, a decoder-only transformer using token-level cross-entropy in the SONAR embedding space, achieves competitive text generation quality without diffusion sampling.
Zhangxuan Gu, Zhengwen Zeng, Zhenyu Xu +21 authors
UI-Venus achieves state-of-the-art performance in UI grounding and navigation tasks using reinforcement fine-tuning and a self-evolving framework that enhances trajectory history alignment and action distribution.
Jason Lee, Jiafei Duan, Haoquan Fang +16 authors
Action Reasoning Models (ARMs) integrate perception, planning, and control to enable adaptable and explainable robotic behavior, achieving superior performance across various tasks and settings.
Zijian Chen, Xueguang Ma, Shengyao Zhuang +17 authors
BrowseComp-Plus, a curated benchmark, enables controlled evaluation of deep research agents and retrieval methods, providing insights into their performance and effectiveness.
Zhenpeng Su, Leiyu Pan, Xue Bai +5 authors
Klear-Reasoner, a model with long reasoning capabilities, achieves high performance across benchmarks through detailed post-training workflows, including long Chain-of-Thought supervised fine-tuning and reinforcement learning with Gradient-Preserving clipping Policy Optimization.
Jiatong Li, Weida Wang, Qinggang Zhang +6 authors
Mol-R1 framework enhances molecule discovery by improving reasoning performance and explainability through PRID and MoIA strategies.
Mian Zhang, Shujian Liu, Sixun Dong +9 authors
LogicIFGen and LogicIFEval assess the instruction-following capabilities of LLMs on complex, logic-rich instructions, revealing significant performance gaps.
Bowen Xue, Qixin Yan, Wenjing Wang +2 authors
A lightweight framework for identity preservation in video generation using conditional image branches and restricted self-attentions outperforms full-parameter methods with minimal additional parameters.
Xiang Zhang, Yawar Siddiqui, Armen Avetisyan +3 authors
VertexRegen generates meshes with continuous detail by reversing edge collapse through a generative model, offering anytime generation and flexibility in detail levels.
Junyao Gao, Jiaxing Li, Wenran Liu +5 authors
CharacterShot is a 4D character animation framework that uses a DiT-based model and dual-attention module to generate consistent 3D animations from a single image and 2D pose sequence.
Wen Wang, Bozhen Fang, Chenchen Jing +6 authors
Two methods, Temporal Self-Consistency Voting and Temporal Consistency Reinforcement, improve diffusion large language models by leveraging temporal consistency in intermediate predictions.
Runnan Fang, Yuan Liang, Xiaobin Wang +6 authors
Agents equipped with a learnable, updatable procedural memory system, Memp, achieve improved performance and efficiency across tasks by distilling past experiences into detailed instructions and higher-level abstractions.
Tianyi Li, Mingda Chen, Bowei Guo +1 authors
Diffusion Language Models offer parallel token generation, reducing inference latency and capturing bidirectional context, and are compared to autoregressive models in various NLP tasks.
Chaoyue Song, Xiu Li, Fan Yang +6 authors
Puppeteer is a framework that automates rigging and animation of 3D models using an auto-regressive transformer, attention-based architecture, and differentiable optimization, outperforming existing methods in accuracy and efficiency.
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