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。
606 篇论文 · 按点赞排序
Suwhan Choi, Jaeyoon Jung, Haebin Seong +7 authors
D2E framework uses desktop interactions to pretrain embodied AI, achieving high success rates in physical manipulation and navigation tasks.
Mingjie Liu, Shizhe Diao, Ximing Lu +5 authors
Prolonged reinforcement learning training (ProRL) uncovers novel reasoning strategies in language models, outperforming base models and suggesting meaningful expansion of reasoning capabilities.
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.
Qiying Yu, Zheng Zhang, Ruofei Zhu +32 authors
The DAPO algorithm, which includes decoupled clip and dynamic sampling policy optimization, enables open-source, high-performance reinforcement learning training for large-scale language models, enhancing reproducibility in the field.
Xuyang Liu, Zichen Wen, Shaobo Wang +13 authors
The focus in AI research is shifting from model-centric to data-centric compression, with token compression identified as key to improving efficiency in handling long-context scenarios.
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.
Sergey Pletenev, Maria Marina, Nikolay Ivanov +6 authors
EverGreenQA, a multilingual QA dataset with evergreen labels, is introduced to benchmark LLMs on temporality encoding and assess their performance through verbalized judgments and uncertainty signals.
Ming Hu, Chenglong Ma, Wei Li +100 authors
Sci-LLMs are evolving through a co-development with scientific data, addressing unique challenges like multimodal and domain-specific information, and are moving towards autonomous, closed-loop systems in scientific research.
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.
HunyuanWorld Team, Zhenwei Wang, Yuhao Liu +52 authors
HunyuanWorld 1.0 generates immersive 3D scenes from text and images using a semantically layered 3D mesh representation with panoramic world proxies, achieving state-of-the-art performance in coherence, exploration, and interactivity.
Kimi Team, Angang Du, Bohong Yin +89 authors
Kimi-VL, an efficient Mixture-of-Experts vision-language model, excels in multimodal reasoning, long-context understanding, and diverse vision-language tasks, achieving competitive performance with reduced computational cost.
Qinsi Wang, Bo Liu, Tianyi Zhou +6 authors
Vision-Zero is a domain-agnostic framework that enhances vision-language models through self-improvement in competitive visual games, using Iterative Self-Play Policy Optimization and achieving state-of-the-art performance without human annotation.
Zonghang Li, Tao Li, Wenjiao Feng +2 authors
Prima.cpp, a distributed inference system, enables running large language models on home devices using CPU/GPU and low memory, overcoming hardware limitations.
Yang Yue, Zhiqi Chen, Rui Lu +5 authors
Despite initial claims, reinforcement learning with verifiable rewards does not introduce fundamentally new reasoning abilities to LLMs, instead it enhances performance by biasing output distribution toward rewarded paths without expanding the reasoning boundary.
Cong Wei, Bo Sun, Haoyu Ma +10 authors
MoCha generates realistic talking character animations from speech and text using a speech-video attention mechanism and joint training on speech-labeled and text-labeled data, enabling multi-character conversations and superior realism.
Xufang Luo, Yuge Zhang, Zhiyuan He +5 authors
Agent Lightning is a flexible RL framework for training LLMs in various agents, using a hierarchical RL algorithm and decoupling execution from training to handle complex interactions.
Sen Xu, Yi Zhou, Wei Wang +7 authors
VibeThinker-1.5B, a 1.5B-parameter model using the Spectrum-to-Signal Principle, achieves superior reasoning capabilities compared to larger models at a significantly lower cost.
Shalini Maiti, Amar Budhiraja, Bhavul Gauri +8 authors
Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse domains, but their training remains resource- and time-intensive, requiring massive compute power and careful orchestration of training procedures. Model souping-the practice of averaging weights from multiple models of the same architecture-has emerged as a promising pre- and post-training technique that can enhance performance without expensive retraining. In this paper, we introduce Soup Of Category Experts (SoCE), a principled approach for model souping that utilizes benchmark composition to identify optimal model candidates and applies non-uniform weighted averaging to maximize performance. Contrary to previous uniform-averaging approaches, our method leverages the observation that benchmark categories often exhibit low inter-correlations in model performance. SoCE identifies "expert" models for each weakly-correlated category cluster and combines them using optimized weighted averaging rather than uniform weights. We demonstrate that the proposed method improves performance and robustness across multiple domains, including multilingual capabilities, tool calling, and math and achieves state-of-the-art results on the Berkeley Function Calling Leaderboard.
Francesco Capuano, Caroline Pascal, Adil Zouitine +2 authors
Robot learning transitions from model-based to data-driven methods, leveraging reinforcement learning and behavioral cloning to develop versatile, language-conditioned models for diverse tasks and robot types.
Yuxuan Song, Zheng Zhang, Cheng Luo +19 authors
Seed Diffusion Preview, a discrete-state diffusion language model, achieves fast inference speeds through parallel generation, outperforming Mercury and Gemini Diffusion in speed and quality.
Bowen Zhang, Congchao Guo, Geng Yang +17 authors
MiniMax-Speech, an autoregressive Transformer-based TTS model, generates high-quality speech with a learnable speaker encoder that extracts reference speaker features without transcription, achieving SOTA results in voice cloning and supporting various extensions.
Runze Zhang, Guoguang Du, Xiaochuan Li +10 authors
The DropletVideo model preserves spatio-temporal coherence in generated videos by considering both camera movements and plot developments, leveraging a large annotated dataset.
Nicolas Carion, Laura Gustafson, Yuan-Ting Hu +35 authors
Segment Anything Model 3 achieves state-of-the-art performance in promptable concept segmentation and tracking by leveraging a unified model architecture with decoupled recognition and localization.
Xu Wujiang, Wentian Zhao, Zhenting Wang +6 authors
Entropy-regularized Policy Optimization (EPO) addresses exploration-exploitation challenges in multi-turn environments with sparse rewards, improving performance in tasks like ScienceWorld and ALFWorld.
Xingxuan Li, Yao Xiao, Dianwen Ng +15 authors
The MiroMind-M1 series, built on Qwen-2.5, introduces fully open-source reasoning language models with state-of-the-art performance in mathematical reasoning through Context-Aware Multi-Stage Policy Optimization.
Chaorui Deng, Deyao Zhu, Kunchang Li +9 authors
BAGEL, an open-source foundational model trained on diverse multimodal data, significantly outperforms existing models in both generation and understanding tasks.
Fei Tang, Zhangxuan Gu, Zhengxi Lu +9 authors
A new reward framework, GUI-G$^2$, models GUI elements as continuous Gaussian distributions to improve autonomous interaction through dense gradient signals, outperforming existing methods in spatial reasoning tasks.
Pengyi Li, Matvey Skripkin, Alexander Zubrey +2 authors
Reinforcement Learning via Self-Confidence (RLSC) improves large language model accuracy using the model's confidence as a reward signal, eliminating the need for human labels or reward engineering.
Jiacheng Chen, Qianjia Cheng, Fangchen Yu +25 authors
Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric. Physics is the sharpest test of this shift, which binds symbols to reality in a fundamental way, serving as the cornerstone of most modern technologies. In this work, we manage to advance physics research by developing large language models with exceptional physics reasoning capabilities, especially excel at solving Olympiad-level physics problems. We introduce P1, a family of open-source physics reasoning models trained entirely through reinforcement learning (RL). Among them, P1-235B-A22B is the first open-source model with Gold-medal performance at the latest International Physics Olympiad (IPhO 2025), and wins 12 gold medals out of 13 international/regional physics competitions in 2024/2025. P1-30B-A3B also surpasses almost all other open-source models on IPhO 2025, getting a silver medal. Further equipped with an agentic framework PhysicsMinions, P1-235B-A22B+PhysicsMinions achieves overall No.1 on IPhO 2025, and obtains the highest average score over the 13 physics competitions. Besides physics, P1 models also present great performance on other reasoning tasks like math and coding, showing the great generalibility of P1 series.
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.
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