Repository navigation
Expand file tree
/
Copy pathprojects.html
More file actions
12 lines (10 loc) · 10.8 KB
/
Copy pathprojects.html
File metadata and controls
12 lines (10 loc) · 10.8 KB
1
2
3
4
5
6
7
8
9
10
11
<!DOCTYPE html>
<html lang="zh-CN"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title data-en="Projects — DIG">项目 — DIG</title><meta name="description" content="北京大学杨灵课题组:大模型与智能体后训练、强化学习系统、递归自我改进和发现智能。" data-en-content="Ling Yang's research group at Peking University: post-training, RL systems, recursive self-improvement, and Discovery Intelligence."><link rel="icon" href="assets/dig-logo.png"><link rel="stylesheet" href="assets/site.css"><link rel="stylesheet" href="assets/compact.css"><link rel="stylesheet" href="assets/bilingual.css"><link rel="stylesheet" href="assets/partners.css"></head><body><nav class="site-nav"><div class="nav-in"><a class="brand" href="index.html"><img src="assets/dig-logo.png" alt="DIG — Discovery Intelligence Group"></a><div class="nav-links" id="site-links"><a data-en="Homepage" href="index.html">首页</a><a data-en="Research" href="research.html">研究方向</a><a data-en="People" href="people.html">成员</a><a data-en="Projects" href="projects.html">项目</a><a data-en="Publications" href="publications.html">论文</a><a class="nav-cta" data-en="Join DIG" href="join.html">招生</a></div><div class="nav-tools"><button class="lang-btn" id="lang-toggle" type="button" aria-label="Switch to English" title="Switch to English">EN</button><button class="menu-btn" type="button" aria-label="展开导航" aria-controls="site-links" aria-expanded="false">☰</button></div></div></nav>
<header class="page-hero"><div class="wrap"><h1 data-en="Projects">项目</h1><p data-en="Papers, code, models, and online demos.">论文、代码、模型与在线演示。</p></div></header><section class="section"><div class="wrap"><div class="project-grid"><div class="project-card"><div class="type" data-en="Benchmark · Discovery Intelligence">评测基准 · 发现智能</div><h3>Science's Last Exam</h3><p data-en="Challenging and measuring AI agents on real scientific research, end to end: six scientific capabilities plus one end-to-end research assessment, 113 tasks across nine disciplines.">在真实科研全流程上挑战并度量 AI 智能体:六项科学能力与一项端到端科研评估,共 113 个任务,覆盖九个学科。</p><div class="project-links"><a data-en="Website ↗" href="https://sciences-last-exam.org/">官网 ↗</a><a data-en="Code ↗" href="https://github.com/Gen-Verse/Science-Last-Exam">代码 ↗</a><a data-en="Data ↗" href="https://huggingface.co/datasets/Gen-Verse/Science-Last-Exam">数据 ↗</a><a data-en="Leaderboard ↗" href="https://sciences-last-exam.org/leaderboard">排行榜 ↗</a></div></div><div class="project-card"><div class="type" data-en="Discovery Intelligence · RSI">发现智能 · 递归自我改进</div><h3>DiscoveryRSI</h3><p data-en="Recursively improving research models and agent harnesses by learning from hypothesis revisions, scientific contributions, and executable investigations.">从科研假设迭代、关键科研贡献与可执行研究中学习,递归改进研究模型与智能体执行框架。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2610.23839">论文 ↗</a><a data-en="Code ↗" href="https://github.com/Gen-Verse/DiscoveryRSI">代码 ↗</a><a data-en="Model ↗" href="https://huggingface.co/Gen-Verse/DiscoveryRSI-27B">模型 ↗</a></div></div><div class="project-card"><div class="type" data-en="Discovery Intelligence">发现智能</div><h3>Discovery Foundation Models</h3><p data-en="Seven core capabilities for open-ended discovery intelligence and the Zetema reference architecture.">提出开放式发现智能的七项核心能力及 Zetema 参考架构。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2609.15973">论文 ↗</a><a data-en="Roadmap ↗" href="https://github.com/Gen-Verse/DFM-Plans">研究路线 ↗</a></div></div><div class="project-card"><div class="type" data-en="Scientific Agents · RSI">科学智能体 · 递归自我改进</div><h3>ScienceBuddy</h3><p data-en="Recursive co-improvement of agent harnesses and models through scientific interaction.">通过科研交互,递归改进智能体执行框架(harness)与模型。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2609.17523">论文 ↗</a><a data-en="Code ↗" href="https://github.com/Gen-Verse/ScienceBuddy">代码 ↗</a><a data-en="Demo ↗" href="http://science-buddy.io/">试用 ↗</a></div></div><div class="project-card"><div class="type" data-en="Scientific Environments">科学学习环境</div><h3>ScienceIDE</h3><p data-en="Scientific code transformed into executable, verifiable environments for agent training and evaluation.">将科学代码转化为可执行、可验证的智能体训练与评测环境。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2609.19134">论文 ↗</a><a data-en="Code ↗" href="https://github.com/aitofound/ScienceIDE">代码 ↗</a><a data-en="Models ↗" href="https://huggingface.co/collections/AItonomy/scienceide-model-series">模型 ↗</a></div></div><div class="project-card"><div class="type" data-en="World Models · OPF">世界模型 · OPF</div><h3>JEPA-Anything</h3><p data-en="Cross-domain predictive models using Orthogonal Predictive Factorization (OPF).">以正交预测因子分解(OPF)学习跨领域的预测模型。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2609.20800">论文 ↗</a><a data-en="Code ↗" href="https://github.com/Gen-Verse/JEPA-Anything">代码 ↗</a><a data-en="Models ↗" href="https://huggingface.co/collections/Gen-Verse/jepa-anything">模型 ↗</a></div></div><div class="project-card"><div class="type" data-en="World Models · RSI">世界模型 · 递归自我改进</div><h3>RSI-Anything</h3><p data-en="Built on JEPA-Anything, using world-model predictions and interaction feedback for recursive self-improvement across domains.">基于 JEPA-Anything,结合世界模型预测与交互反馈,实现跨领域递归自我改进。</p><div class="project-links"><a data-en="Paper ↗" href="https://arxiv.org/abs/2610.13912">论文 ↗</a><a data-en="Code ↗" href="https://github.com/Gen-Verse/RSI-Anything">代码 ↗</a></div></div><div class="project-card"><div class="type" data-en="Long-Horizon Agents · Memory">长程智能体 · 记忆进化</div><h3>Recuris</h3><p data-en="Couples experiential and working memory, using environment feedback to refine skills and execution in long-horizon tasks.">耦合经验记忆与工作记忆,利用环境反馈改进长程任务中的技能和执行过程。</p><div class="project-links"><a href="https://github.com/Gen-Verse/Recuris">GitHub ↗</a><a data-en="Paper ↗" href="https://arxiv.org/abs/2608.24876">论文 ↗</a></div></div><div class="project-card"><div class="type" data-en="Reasoning · Process Rewards">推理与过程奖励</div><h3>ReasonFlux</h3><p data-en="Reasoning and post-training with thought templates, process reward models, and code generation.">围绕思维模板、过程奖励模型及代码生成开展推理与后训练研究。</p><div class="project-links"><a href="https://github.com/Gen-Verse/ReasonFlux">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Agentic Reinforcement Learning">智能体强化学习</div><h3>RLAnything</h3><p data-en="Dynamic co-optimization of environments, policies, and reward models.">研究环境、策略与奖励模型的动态协同优化。</p><div class="project-links"><a href="https://github.com/Gen-Verse/Open-AgentRL">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Interactive Agent RL">交互式智能体强化学习</div><h3>OpenClaw-RL</h3><p data-en="Trains agents from interaction feedback across conversation, terminals, GUIs, software engineering, and tool use.">利用交互反馈训练智能体,支持对话、终端、GUI、软件工程与工具调用。</p><div class="project-links"><a href="https://github.com/Gen-Verse/OpenClaw-RL">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Environment Construction">环境构建与进化</div><h3>GenEnv</h3><p data-en="Environment construction, curriculum adaptation, and environment–policy co-evolution for agent learning.">研究智能体学习中的环境构建、课程调整与环境—策略协同进化。</p><div class="project-links"><a href="https://github.com/Gen-Verse/GenEnv">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Multimodal Diffusion LMs">多模态扩散语言模型</div><h3>MMaDA</h3><p data-en="Unified multimodal diffusion language models.">研究统一的多模态扩散语言模型。</p><div class="project-links"><a href="https://github.com/Gen-Verse/MMaDA">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Diffusion LLM RL">扩散语言模型强化学习</div><h3>dLLM-RL</h3><p data-en="Trajectory-aware reinforcement learning and post-training for diffusion language models.">面向扩散语言模型的轨迹感知强化学习与后训练。</p><div class="project-links"><a href="https://github.com/Gen-Verse/dLLM-RL">GitHub ↗</a></div></div><div class="project-card"><div class="type" data-en="Multi-Agent Collaboration">多智能体协作</div><h3>LatentMAS</h3><p data-en="Multi-agent reasoning and collaboration in latent space.">研究多智能体在隐空间中的推理与协作。</p><div class="project-links"><a href="https://github.com/Gen-Verse/LatentMAS">GitHub ↗</a></div></div></div></div></section>
<section class="partners" aria-labelledby="partners-title"><div class="wrap"><h2 class="partners-heading" id="partners-title" data-en="Collaborating Institutions">合作单位</h2><ul class="partners-grid">
<li class="partner"><span class="partner-wordmark">Princeton</span><span class="partner-name" data-en="Princeton University">普林斯顿大学</span></li>
<li class="partner"><span class="partner-wordmark">Stanford</span><span class="partner-name" data-en="Stanford University">斯坦福大学</span></li>
<li class="partner"><span class="partner-wordmark">Oxford</span><span class="partner-name" data-en="University of Oxford">牛津大学</span></li>
<li class="partner"><span class="partner-wordmark">PhAI Labs</span><span class="partner-name" data-en="PhAI Labs">未明新域</span></li>
</ul></div></section>
<footer><div class="wrap footer-in"><span data-en="© 2026 Discovery Intelligence Group · Peking University">© 2026 发现智能课题组 · 北京大学</span><span class="footer-links"><a href="https://github.com/Gen-Verse">GitHub</a><a href="https://huggingface.co/Gen-Verse">Hugging Face</a><a data-en="Contact" href="mailto:yangling0818@163.com">联系</a></span></div></footer><script src="assets/site.js?v=20260925-zh1"></script></body></html>