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。
Alexia Jolicoeur-Martineau
Tiny Recursive Model (TRM) achieves high generalization on complex puzzle tasks using a small, two-layer network with minimal parameters, outperforming larger language models.
50 篇论文 · 按点赞排序
Kai Zhang, Xiangchao Chen, Bo Liu +27 authors
Early experience, using agent-generated interaction data without reward signals, improves policy effectiveness and generalization, serving as a bridge between imitation learning and reinforcement learning.
Qizheng Zhang, Changran Hu, Shubhangi Upasani +10 authors
ACE, a framework for adaptive context engineering, enhances LLM applications by preserving detailed knowledge through structured updates, outperforming baselines in agent and domain-specific tasks with reduced adaptation costs.
Shruthan Radhakrishna, Aman Tiwari, Aanjaneya Shukla +21 authors
A 15-billion parameter multimodal reasoning model achieves competitive performance through a progressive training methodology without reinforcement learning, demonstrating efficient use of computational resources.
Zeyu Zhu, Kevin Qinghong Lin, Mike Zheng Shou
PaperTalker is a multi-agent framework that automates academic presentation video generation by integrating slide generation, layout refinement, subtitling, speech synthesis, and talking-head rendering, outperforming existing methods.
Zhuofeng Li, Haoxiang Zhang, Seungju Han +6 authors
AgentFlow, a trainable agentic framework with in-the-flow optimization, enhances reasoning in large language models by coordinating specialized modules and outperforms top baselines across various tasks.
Xiangyu Zhao, Junming Lin, Tianhao Liang +11 authors
Existing Multimodal Large Language Models show performance deficits in long-chain reflective reasoning, which is addressed by developing MM-HELIX-100K and Adaptive Hybrid Policy Optimization, leading to improved accuracy and generalization.
Tianyu Fu, Zihan Min, Hanling Zhang +4 authors
Cache-to-Cache (C2C) enables direct semantic communication between LLMs using neural network projections, improving accuracy and reducing latency compared to text-based communication.
Cong Wei, Quande Liu, Zixuan Ye +5 authors
UniVideo, a dual-stream framework combining a Multimodal Large Language Model and a Multimodal DiT, extends unified modeling to video generation and editing, achieving state-of-the-art performance and supporting task composition and generalization.
Shreyas Singh, Kunal Singh, Pradeep Moturi
Fathom-DeepResearch, an agentic system with specialized models for web search and report synthesis, achieves state-of-the-art performance on open-ended information-seeking tasks and diverse reasoning tasks.
Ziyuan Huang, DanDan Zheng, Cheng Zou +13 authors
MingTok, a continuous latent space visual tokenizer, unifies vision-language understanding and generation within an autoregressive framework, achieving state-of-the-art performance across both domains.
Guanhua Huang, Tingqiang Xu, Mingze Wang +7 authors
Low-probability Regularization (Lp-Reg) enhances exploration in Reinforcement Learning with Verifiable Rewards (RLVR) by preserving valuable low-probability tokens, leading to improved performance in complex reasoning tasks.
Bin Xia, Bohao Peng, Yuechen Zhang +10 authors
DreamOmni2 addresses limitations in instruction-based image editing and subject-driven generation by introducing multimodal instruction-based editing and generation tasks, utilizing feature mixing, index encoding, and joint training with a VLM.
Youjin Wang, Yangjingyi Chen, Jiahao Yan +2 authors
MemMamba, a novel architecture integrating state summarization and cross-attention, improves long-range memory and efficiency in sequence modeling compared to Mamba and Transformers.
Jiaru Zou, Soumya Roy, Vinay Kumar Verma +6 authors
TaTToo, a novel table-grounded Process Reward Model, enhances tabular reasoning by explicitly addressing table-specific operations and integrating tool-based verification, leading to significant performance improvements over existing PRMs.
Minghong Cai, Qiulin Wang, Zongli Ye +7 authors
VideoCanvas addresses temporal ambiguity in latent video diffusion models to enable flexible spatio-temporal video completion using a hybrid conditioning strategy.
ShengYun Peng, Eric Smith, Ivan Evtimov +7 authors
RECAP, a reinforcement learning method, enhances the safety and robustness of large reasoning models by teaching them to override flawed reasoning and maintain safety without additional training costs.
Chengyue Wu, Hao Zhang, Shuchen Xue +7 authors
Fast-dLLM v2, a block diffusion language model, efficiently converts pretrained autoregressive models for parallel text generation, achieving significant speedup without compromising accuracy.
Yoonjeon Kim, Doohyuk Jang, Eunho Yang
A training pipeline called MASA enhances meta-awareness in reasoning models, leading to improved accuracy and efficiency across various benchmarks.
Yi Xin, Qi Qin, Siqi Luo +29 authors
Lumina-DiMOO, an open-source foundational model, uses fully discrete diffusion modeling for efficient multi-modal generation and understanding, outperforming existing models.
Yunlong Tang, Jing Bi, Pinxin Liu +24 authors
This survey examines post-training methodologies for Video-LMMs, focusing on supervised fine-tuning, reinforcement learning, and test-time scaling, while addressing challenges in video understanding.
Soyeong Jeong, Taehee Jung, Sung Ju Hwang +2 authors
Thought templates enhance long-context language models by structuring evidence combination and guiding multi-hop inference, leading to consistent performance improvements across various benchmarks.
Cheng Yang, Jiaxuan Lu, Haiyuan Wan +2 authors
ChemMAS, a multi-agent system, improves reaction condition recommendation by providing interpretable rationales, outperforming existing methods in accuracy and explainability.
Yuzheng Cai, Siqi Cai, Yuchen Shi +10 authors
Training-Free GRPO enhances LLM agent performance in specialized domains by learning experiential knowledge as a token prior without parameter updates, improving out-of-domain tasks with minimal data.
Hongzhi Zang, Mingjie Wei, Si Xu +14 authors
RLinf-VLA is a unified framework for scalable reinforcement learning training of vision-language-action models, offering improved performance and generalization compared to supervised fine-tuning.
Haolin Chen, Shiyu Wang, Can Qin +12 authors
CoDA, a 1.7B-parameter diffusion coder, achieves competitive performance with smaller models through confidence-guided sampling and is released with open-source tools.
Jiaxi Li, Yucheng Shi, Jin Lu +1 authors
Mutual Information Tree Search (MITS) uses information-theoretic principles to guide and evaluate reasoning paths in large language models, improving performance and efficiency.
Jingyu Zhang, Haozhu Wang, Eric Michael Smith +7 authors
WaltzRL, a multi-agent reinforcement learning framework, improves LLM safety and helpfulness by collaboratively training a conversation agent and a feedback agent, reducing unsafe responses and overrefusals.
Zichen Wen, Shaobo Wang, Yufa Zhou +8 authors
EPIC, a progressive learning framework, improves the efficiency of multi-modal large models by reducing training difficulty through token and layer consistency distillation during visual token compression.
Ziqi Huang, Ning Yu, Gordon Chen +3 authors
VChain enhances video generation by integrating visual reasoning from multimodal models to guide sparse tuning of a pre-trained video generator.
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