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。
Jeffrey Amico, Gabriel Passamani Andrade, John Donaghy +12 authors
Swarm sAmpling Policy Optimization (SAPO) is a decentralized and asynchronous RL algorithm that enhances post-training language models without supervised fine-tuning, achieving significant reward gains and scalability across diverse hardware.
50 篇论文 · 按点赞排序
Yihao Wang, Pengxiang Ding, Lingxiao Li +13 authors
VLA-Adapter reduces reliance on large-scale VLMs and extensive pre-training by using a lightweight Policy module with Bridge Attention, achieving state-of-the-art performance and fast inference speed with minimal computational resources.
Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala +1 authors
Language models produce incorrect statements due to training and evaluation procedures that reward guessing over acknowledging uncertainty, leading to a need for socio-technical changes in benchmark scoring.
Kaiyan Zhang, Yuxin Zuo, Bingxiang He +36 authors
Reinforcement Learning enhances Large Language Models for complex reasoning tasks, facing challenges in scalability and infrastructure as the field advances.
Haozhe Wang, Haoran Que, Qixin Xu +9 authors
REER, a new paradigm for deep reasoning, uses reverse engineering to discover step-by-step reasoning processes, enabling a model to perform competitively on open-ended tasks.
Liyang Chen, Tianxiang Ma, Jiawei Liu +7 authors
HuMo is a unified framework for human-centric video generation that addresses challenges in multimodal control through a two-stage training paradigm and novel strategies for subject preservation and audio-visual synchronization.
Tong Zheng, Hongming Zhang, Wenhao Yu +8 authors
Parallel-R1, a reinforcement learning framework, enhances large language models' reasoning capabilities by enabling parallel thinking through a progressive curriculum, leading to significant performance improvements on math benchmarks.
Heeji Yoon, Jaewoo Jung, Junwan Kim +10 authors
VIRAL, a regularization strategy, aligns MLLMs' visual representations with pre-trained VFMs, enhancing performance on vision-centric tasks.
Junteng Liu, Yunji Li, Chi Zhang +12 authors
WebExplorer, a data-driven approach for developing advanced web agents, achieves state-of-the-art performance in information-seeking tasks through systematic data generation and reinforcement learning.
Haozhan Li, Yuxin Zuo, Jiale Yu +18 authors
SimpleVLA-RL, an RL framework for VLA models, enhances long-horizon action planning, achieves state-of-the-art performance, and discovers novel patterns during training.
Jie Wu, Yu Gao, Zilyu Ye +9 authors
RewardDance is a scalable reward modeling framework that aligns with VLM architectures, enabling effective scaling of RMs and resolving reward hacking issues in generation models.
Haoyu Dong, Pengkun Zhang, Mingzhe Lu +2 authors
MachineLearningLM enhances a general-purpose LLM with robust in-context machine learning capabilities through continued pretraining with synthesized ML tasks, achieving high performance across various domains without task-specific training.
Yuhao Zhang, Yuhao Du, Zhanchen Dai +4 authors
EchoX, a speech-to-speech large language model, addresses the acoustic-semantic gap by integrating semantic representations, preserving reasoning abilities, and achieving advanced performance on knowledge-based benchmarks.
Xin Lai, Junyi Li, Wei Li +3 authors
Mini-o3, a system for deep, multi-turn reasoning in visual search tasks, uses an iterative data collection pipeline and over-turn masking strategy to achieve state-of-the-art performance with rich reasoning patterns.
Lingdong Kong, Wesley Yang, Jianbiao Mei +20 authors
This survey provides a comprehensive review of 3D and 4D world modeling and generation, establishing definitions, taxonomy, datasets, and evaluation metrics, and discussing applications and challenges.
Yinjie Wang, Ling Yang, Bowen Li +3 authors
TraceRL enhances diffusion language models with trajectory-aware reinforcement learning, improving reasoning performance on complex tasks and enabling flexible sampling.
Zhiheng Xi, Jixuan Huang, Chenyang Liao +20 authors
AgentGym-RL is a modular RL framework for training LLM agents in diverse environments without supervised fine-tuning, featuring ScalingInter-RL for balanced exploration-exploitation.
Itai Gat, Heli Ben-Hamu, Marton Havasi +6 authors
Set Block Decoding accelerates language model generation by integrating next token prediction and masked token prediction, enabling parallel sampling of future tokens and reducing computational cost without sacrificing accuracy.
Yikang Ding, Jiwen Liu, Wenyuan Zhang +11 authors
Kling-Avatar, a cascaded framework, enhances audio-driven avatar video generation by integrating multimodal instruction understanding with photorealistic portrait generation, resulting in high-fidelity, semantically grounded videos.
Jiawei Wang, Jiacai Liu, Yuqian Fu +7 authors
Entropy-Modulated Policy Gradients (EMPG) addresses learning dynamics issues in LLMs by recalibrating policy gradients based on uncertainty and task outcomes, leading to improved performance in long-horizon tasks.
Yamei Chen, Haoquan Zhang, Yangyi Huang +4 authors
LLMs generate SVGs from natural-language descriptions using a reinforcement learning approach with verifiable rewards, improving performance and scene coherence.
Rongyao Fang, Aldrich Yu, Chengqi Duan +7 authors
FLUX-Reason-6M and PRISM-Bench address the lack of reasoning-focused datasets and benchmarks for text-to-image models, providing a large-scale dataset and evaluation standard to improve model performance.
Jiacheng Miao, Joe R. Davis, Jonathan K. Pritchard +1 authors
Paper2Agent converts research papers into interactive AI agents to facilitate knowledge dissemination and enable complex scientific queries through natural language.
Ji Xie, Trevor Darrell, Luke Zettlemoyer +1 authors
Reconstruction Alignment (RecA) is a post-training method that enhances multimodal models by using visual embeddings as dense prompts, improving image generation and editing fidelity.
Che Liu, Yinda Chen, Haoyuan Shi +16 authors
DINOv3, a self-supervised vision transformer, demonstrates strong performance across various medical vision tasks without domain-specific pre-training, though it shows limitations in deeply specialized domains and does not consistently follow scaling laws in the medical domain.
Jiahao Wang, Yufeng Yuan, Rujie Zheng +12 authors
SpatialVID, a large-scale dataset with diverse videos and dense 3D annotations, enhances model generalization and performance in video and 3D vision research.
Zhiyuan Yan, Kaiqing Lin, Zongjian Li +11 authors
A novel framework UAE uses reinforcement learning to unify image-to-text and text-to-image processes, enhancing mutual understanding and generation fidelity.
Qi Lv, Weijie Kong, Hao Li +7 authors
F1, a pretrained VLA framework with foresight generation, improves task success and generalization in dynamic environments through a Mixture-of-Transformer architecture and next-scale prediction.
Haozhe Wang, Qixin Xu, Che Liu +3 authors
Reinforcement Learning enhances LLM reasoning through a two-phase process involving procedural correctness and strategic planning, with HICRA algorithm focusing on high-impact planning tokens to improve performance.
Jakub Grudzien Kuba, Mengting Gu, Qi Ma +2 authors
Language Self-Play (LSP) enhances large language models' performance on instruction-following tasks through self-play, surpassing data-driven methods.
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