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。
50 篇论文 · 按点赞排序
Hao Wang, Limeng Qiao, Chi Zhang +4 authors
X2SAM is a unified multimodal model that extends segmentation capabilities from images to videos while supporting conversational instructions and visual prompts for both modalities.
Yian Zhao, Feng Wang, Qiushan Guo +4 authors
Predictive Video VAE combines predictive learning with video reconstruction to improve latent space representation and generative performance through temporal coherence and motion priors.
Jianing Wang, Linsen Guo, Zhengyu Chen +8 authors
HeavySkill presents a framework where complex reasoning is internalized as an intrinsic model skill rather than relying on external orchestration, demonstrating superior performance through parallel reasoning and summarization stages that can be enhanced via reinforcement learning.
Gal Yona, Mor Geva, Yossi Matias
Generative AI models continue to produce factual errors despite improvements in reliability, with the key challenge being distinguishing known from unknown information rather than simply expanding knowledge boundaries.
Cheng Qian, Hyeonjeong Ha, Jiayu Liu +10 authors
Large language models demonstrate limited creative problem-solving abilities when required to repurpose objects based on affordance reasoning, indicating a gap in current AI capabilities for novel tool usage.
Qiliang Liang, Hansi Wang, Zhong Liang +1 authors
Structured representation of agent skills disentangles scheduling, execution, and logic components, improving performance in skill discovery and risk assessment tasks.
Ansar Aynetdinov, Patrick Haller, Alan Akbik
Training high-quality filtered data multiple times yields better performance than single-pass training on larger, less filtered datasets for non-English language models.
Shitong Shao, Zikai Zhou, Haopeng Li +4 authors
In-context sparse attention framework enables efficient video editing with reduced computational costs while maintaining visual quality.
Lin Song, Wenbo Li, Guoqing Ma +16 authors
JoyAI-Image integrates a spatially enhanced MLLM with MMDiT to achieve unified visual understanding, text-to-image generation, and instruction-guided image editing with enhanced spatial intelligence.
Etai Sella, Hao Phung, Nitay Amiel +3 authors
A training-free framework for fine-grained 3D editing that uses geometric primitives and vision-language models to preserve identity while enabling localized structural changes.
Daniel Zheng, Ingrid von Glehn, Yori Zwols +15 authors
We introduce the AI co-mathematician, a workbench for mathematicians to interactively leverage AI agents to pursue open-ended research. The AI co-mathematician is optimized to provide holistic support for the exploratory and iterative reality of mathematical workflows, including ideation, literature search, computational exploration, theorem proving and theory building. By providing an asynchronous, stateful workspace that manages uncertainty, refines user intent, tracks failed hypotheses, and outputs native mathematical artifacts, the system mirrors human collaborative workflows. In early tests, the AI co-mathematician helped researchers solve open problems, identify new research directions, and uncover overlooked literature references. Besides demonstrating a highly interactive paradigm for AI-assisted mathematical discovery, the AI co-mathematician also achieves state of the art results on hard problem-solving benchmarks, including scoring 48% on FrontierMath Tier 4, a new high score among all AI systems evaluated.
Tianle Wang, Zhaoyang Wang, Guangchen Lan +4 authors
ScaleLogic demonstrates that reinforcement learning training compute scales as a power law with reasoning depth, with scaling exponents increasing monotonically with logical expressiveness across multiple reasoning tasks.
Jingjie Ning, Xiaochuan Li, Ji Zeng +2 authors
Auto research operates as an empirical loop where agents iteratively refine code based on evaluation feedback, achieving improved performance across multiple tasks without human intervention.
Junjie Yu, Pengrui Lu, Weiye Si +75 authors
AcademiClaw presents a comprehensive benchmark for evaluating AI agents on complex academic tasks spanning multiple domains, revealing significant capability gaps in current models.
Chengshuai Shi, Wenzhe Li, Xinran Liang +10 authors
Given the rapidly growing capabilities of vision-language models (VLMs), extending them to interactive decision-making tasks such as video games has emerged as a promising frontier. However, existing approaches either rely on large-scale supervised fine-tuning (SFT) on human trajectories or apply reinforcement learning (RL) only in relatively short-horizon settings (typically around 20--30 turns). In this work, we study RL-based training of VLMs for long-horizon decision-making in Super Mario Land, a visually grounded environment requiring 100+ turns of interaction with coordinated perception, reasoning, and action. We begin with a systematic investigation of key algorithmic components and propose an adapted variant of PPO with a lightweight turn-level critic, which substantially improves training stability and sample efficiency over critic-free methods such as GRPO and Reinforce++. We further show that pretrained VLMs provide strong action priors, significantly improving sample efficiency during RL training and reducing the need for manual design choices such as action engineering, compared to classical deep RL trained from scratch. Building on these insights, we introduce Odysseus, an open training framework for VLM agents, achieving substantial gains across multiple levels of the game and at least 3 times average game progresses than frontier models. Moreover, the trained models exhibit consistent improvements under both in-game and cross-game generalization settings, while maintaining general-domain capabilities. Overall, our results identify key ingredients for making RL stable and effective in long-horizon, multi-modal settings, and provide practical guidance for developing VLMs as embodied agents.
Minchan Kwon, Sunghyun Baek, Minseo Kim +3 authors
Stable-GFN addresses training instability and mode collapse in generative flow networks for large language model red-teaming through partition function elimination and robust masking techniques.
Jinchao Li, Yunxin Li, Chenrui Zhao +3 authors
A cross-application workflow benchmark named WindowsWorld was developed to evaluate GUI agents on complex multi-step tasks requiring coordination across multiple software applications, revealing significant performance gaps in current agents when handling real-world professional workflows.
Yida Xue, Ningyu Zhang, Tingwei Wu +5 authors
OceanPile presents a large-scale multimodal corpus for ocean science, combining diverse data types and a knowledge graph-guided instruction dataset to advance marine AI applications.
Dingwei Chen, Zefang Zong, Zhipeng Ma +5 authors
Reinforcement learning for agentic LLMs suffers from sparse rewards and challenges in credit assignment, which are addressed through A²TGPO that adapts information gain normalization, accumulation, and clipping for improved policy optimization.
Yi Wang, Xinchen Li, Pengwei Xie +13 authors
Learning While Deploying framework enables continuous improvement of Vision-Language-Action policies through fleet-scale offline-to-online reinforcement learning with distributed robot experience and human interventions.
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