[{"data":1,"prerenderedAt":369},["ShallowReactive",2],{"content-query-ylaSQeG0vg":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":363,"_id":364,"_source":365,"_file":366,"_stem":367,"_extension":368},"/technology-blogs/zh/3099","zh",false,"","论文精讲 | 基于昇思MindSpore的多粒度因果结构学习有效提升因果关系发现的准确率","作者：李锐锋 ｜来源：知乎","2024-04-26","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/2fc89b0678594bce9b164160e5324ca7.png","technology-blogs",{"type":14,"children":15,"toc":360},"root",[16,24,43,48,53,58,63,68,79,84,93,98,103,111,119,127,132,137,145,153,165,177,185,193,198,205,210,215,223,231,236,243,248,253,260,265,270,277,282,287,295,303,308,313,318,326],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"论文精讲-基于昇思mindspore的多粒度因果结构学习有效提升因果关系发现的准确率",[22],{"type":23,"value":8},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28,30,36,38],{"type":23,"value":29},"**作者：**",{"type":17,"tag":31,"props":32,"children":33},"strong",{},[34],{"type":23,"value":35},"李锐锋",{"type":23,"value":37}," ｜",{"type":17,"tag":31,"props":39,"children":40},{},[41],{"type":23,"value":42},"来源：知乎",{"type":17,"tag":25,"props":44,"children":45},{},[46],{"type":23,"value":47},"论文标题",{"type":17,"tag":25,"props":49,"children":50},{},[51],{"type":23,"value":52},"Multi-Granularity Causal Structure Learning",{"type":17,"tag":25,"props":54,"children":55},{},[56],{"type":23,"value":57},"论文来源",{"type":17,"tag":25,"props":59,"children":60},{},[61],{"type":23,"value":62},"AAAI 2024",{"type":17,"tag":25,"props":64,"children":65},{},[66],{"type":23,"value":67},"论文链接",{"type":17,"tag":25,"props":69,"children":70},{},[71],{"type":17,"tag":72,"props":73,"children":77},"a",{"href":74,"rel":75},"https://ojs.aaai.org/index.php/AAAI/article/view/29278",[76],"nofollow",[78],{"type":23,"value":74},{"type":17,"tag":25,"props":80,"children":81},{},[82],{"type":23,"value":83},"代码链接",{"type":17,"tag":25,"props":85,"children":86},{},[87],{"type":17,"tag":72,"props":88,"children":91},{"href":89,"rel":90},"http://www.sdu-idea.cn/codes.php?name=MgCSL",[76],[92],{"type":23,"value":89},{"type":17,"tag":25,"props":94,"children":95},{},[96],{"type":23,"value":97},"昇思MindSpore作为开源的AI框架，为产学研和开发人员带来端边云全场景协同、极简开发、极致性能、安全可信的体验，支持超大规模AI预训练，自2020年3月28日开源以来已超过700W+的下载量。昇思MindSpore已支持上千篇AI顶会论文，走入290+所高校进行教学，通过HMS在5000+App上商用，拥有数量众多的开发者，在AI计算中心、智能制造、金融、云、无线、数通、能源、消费者1+8+N、智能汽车等端边云车全场景广泛应用，是Gitee指数最高的开源软件。欢迎大家参与开源贡献、套件、模型众智、行业创新与应用、算法创新、学术合作、AI书籍合作等，贡献您在云侧、端侧、边侧以及安全领域的应用案例。",{"type":17,"tag":25,"props":99,"children":100},{},[101],{"type":23,"value":102},"在科技界、学术界和工业界对昇思MindSpore的广泛支持下，基于昇思MindSpore的AI论文2023年在所有AI框架中占比8%，连续两年进入全球第二，感谢CAAI和各位高校老师支持，我们一起继续努力做好AI科研创新。昇思MindSpore社区支持顶级会议论文研究，持续构建原创AI成果。我会不定期挑选一些优秀的论文来推送和解读，希望更多的产学研专家跟昇思MindSpore合作，一起推动原创AI研究，昇思MindSpore社区会持续支撑好AI创新和AI应用，本文是昇思MindSpore AI顶会论文系列第32篇，我选择了来自山东大学软件学院的余国先和王峻老师团队发表于AAAI2024的一篇论文解读，感谢各位专家教授同学的投稿，更多精彩的论文精读文章和开源代码实现请扫下方二维码访问Models。",{"type":17,"tag":25,"props":104,"children":105},{},[106],{"type":17,"tag":107,"props":108,"children":110},"img",{"alt":7,"src":109},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/1e1416979d0f45d581275d9653c4b1d7.png",[],{"type":17,"tag":25,"props":112,"children":113},{},[114],{"type":17,"tag":31,"props":115,"children":116},{},[117],{"type":23,"value":118},"01",{"type":17,"tag":25,"props":120,"children":121},{},[122],{"type":17,"tag":31,"props":123,"children":124},{},[125],{"type":23,"value":126},"研究背景",{"type":17,"tag":25,"props":128,"children":129},{},[130],{"type":23,"value":131},"在数字化转型和科技创新加速的背景下，数据科学领域正经历着从数据中心范式向科学中心范式的重大转变，伴随而来的是在各个研究领域内蔓延开来的因果革命。与依赖于表面相关性的传统分析方法不同，因果结构学习的核心目标是从观测数据中识别变量间的因果关系，并挖掘表征因果结构的有向无环图。在诸如推荐系统、医学诊断、流行病学等领域中，因果结构学习对于稳定推断和合理决策起到至关重要的作用。",{"type":17,"tag":25,"props":133,"children":134},{},[135],{"type":23,"value":136},"鉴于因果结构学习的重要性，已有大量学者投入到相关研究中，提出了基于约束、基于分数和基于梯度的方法，其中基于梯度的方法备受关注。该方法利用可微的有向无环约束，将结构学习问题转换成连续优化问题，并结合机器学习技术搜索有向无环图，有效提高了因果关系发现的准确率。",{"type":17,"tag":25,"props":138,"children":139},{},[140],{"type":17,"tag":31,"props":141,"children":142},{},[143],{"type":23,"value":144},"02",{"type":17,"tag":25,"props":146,"children":147},{},[148],{"type":17,"tag":31,"props":149,"children":150},{},[151],{"type":23,"value":152},"团队介绍",{"type":17,"tag":25,"props":154,"children":155},{},[156,158,163],{"type":23,"value":157},"论文第一作者",{"type":17,"tag":31,"props":159,"children":160},{},[161],{"type":23,"value":162},"梁嘉旋",{"type":23,"value":164},"目前就读于山东大学软件学院（2021-至今），硕士研究生，主要研究方向为因果学习。",{"type":17,"tag":25,"props":166,"children":167},{},[168,170,175],{"type":23,"value":169},"第一作者导师",{"type":17,"tag":31,"props":171,"children":172},{},[173],{"type":23,"value":174},"王峻",{"type":23,"value":176},"，山东大学-南洋理工大学人工智能国际联合研究院教授，博士生导师，长期致力于可信可解释人工智能理论及其在生物医学大数据分析中的应用，主持（完成）4项国家自然科学基金和多项省部级科研项目，作为主要完成人获重庆市科技进步和自然科学奖各1项，山东计算机学会自然科学一等奖1项，以第一（通讯）作者在IEEE TKDE/TNNLS/TCYB、中国科学-信息科学、计算机学报、AAAI、 IJCAI等国内外重要期刊和会议学术发表论文100余篇。研究成果受到国内外信息和生命科学等不同学科领域同行的广泛关注和引用。",{"type":17,"tag":25,"props":178,"children":179},{},[180],{"type":17,"tag":31,"props":181,"children":182},{},[183],{"type":23,"value":184},"03",{"type":17,"tag":25,"props":186,"children":187},{},[188],{"type":17,"tag":31,"props":189,"children":190},{},[191],{"type":23,"value":192},"论文简介",{"type":17,"tag":25,"props":194,"children":195},{},[196],{"type":23,"value":197},"本文主要针对多粒度因果结构识别困难的问题，着眼于改进因果方法的多粒度因果关系学习能力。现有方法通常独立地考虑单个变量（微观变量）间的因果作用，忽略了多个变量（宏观变量）涌现出的复杂协同作用及其共同行为模式，阻碍了因果结构的正确识别。例如，大脑可以表征为微观层面的神经元及突触，但高阶的协同子系统普遍存在，它们通常位于典型的功能网络之间，并可能出现协同作用。然而，宏观变量的未知性增加了提取有效粗粒化策略进而发现宏观变量的难度。同时，多粒度变量之间的因果方向识别是另一个亟待解决的难题。此外，因果算法的高复杂性使其在处理高维数据时效率显著下降，限制了它们的实际应用。",{"type":17,"tag":25,"props":199,"children":200},{},[201],{"type":17,"tag":107,"props":202,"children":204},{"alt":7,"src":203},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/790106b4b3904ead8c7d6b5534ca8295.png",[],{"type":17,"tag":25,"props":206,"children":207},{},[208],{"type":23,"value":209},"图1 算法框架图",{"type":17,"tag":25,"props":211,"children":212},{},[213],{"type":23,"value":214},"为此，本文提出了一种多粒度因果结构学习算法MgCSL。如图1所示，该算法首先构建一个编码独立的稀疏自编码器来探索有效的粗粒化策略并提取因果变量抽象，其中将编码表示求和以获得潜在的宏观变量表示，并由解码器重构观测数据。从微观变量到宏观变量的贡献矩阵可由编码器参数的路径乘积中提取得到。接下来，MgCSL为每个微观变量构建一个多层感知器，以微观变量和宏观表示作为输入探索潜在的因果机制。为了进一步提高效率，MgCSL引入简化无环约束，以确定多粒度变量之间的因果方向并发现多粒度因果结构。",{"type":17,"tag":25,"props":216,"children":217},{},[218],{"type":17,"tag":31,"props":219,"children":220},{},[221],{"type":23,"value":222},"04",{"type":17,"tag":25,"props":224,"children":225},{},[226],{"type":17,"tag":31,"props":227,"children":228},{},[229],{"type":23,"value":230},"实验结果",{"type":17,"tag":25,"props":232,"children":233},{},[234],{"type":23,"value":235},"为了验证MgCSL的有效性，本文在合成和真实数据集上进行了实验。",{"type":17,"tag":25,"props":237,"children":238},{},[239],{"type":17,"tag":107,"props":240,"children":242},{"alt":7,"src":241},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/d95eb59baa9a4130a2e585ad7077633f.png",[],{"type":17,"tag":25,"props":244,"children":245},{},[246],{"type":23,"value":247},"图 2 多粒度图数据集上的实验结果",{"type":17,"tag":25,"props":249,"children":250},{},[251],{"type":23,"value":252},"首先在多粒度合成数据集上的实验结果如图2所示，宏观变量的存在对基线的准确率产生了影响。即使在小图上，它们的准确率也低至0.5甚至更低。MgCSL能够从多粒度变量的数据中提取有价值的信息用于有向无环图学习，从而以较少的时间开销在准确率和SHD方面实现了最佳性能。",{"type":17,"tag":25,"props":254,"children":255},{},[256],{"type":17,"tag":107,"props":257,"children":259},{"alt":7,"src":258},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/f69f89a1f8374b55b463463cdbe5ac58.png",[],{"type":17,"tag":25,"props":261,"children":262},{},[263],{"type":23,"value":264},"表 1 加性高斯过程的非线性模型数据集上的实验结果",{"type":17,"tag":25,"props":266,"children":267},{},[268],{"type":23,"value":269},"本文也在典型的因果发现合成数据集上进行了实验。如表1所示，MgCSL在大多数指标上都超过了基线，即使在高维图上，其准确率仍然很高。此外，简化的无环惩罚使MgCSL能够在更短的时间内提供有效的结果，确保其在实际应用中的可行性。",{"type":17,"tag":25,"props":271,"children":272},{},[273],{"type":17,"tag":107,"props":274,"children":276},{"alt":7,"src":275},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2024/05/10/0c09d9573d584e3fae2d4ea3a1e86f5c.png",[],{"type":17,"tag":25,"props":278,"children":279},{},[280],{"type":23,"value":281},"表 2 Sachs蛋白质信号数据集上的实验结果",{"type":17,"tag":25,"props":283,"children":284},{},[285],{"type":23,"value":286},"为了进一步验证MgCSL的有效性，本文在Sachs的蛋白质信号数据集上进行了实验。表2中的结果显示，与基线相比，MgCSL仍然表现出了具有竞争力的性能，在准确率、F1 和 SHD 方面均超过了基线，并识别出了6条正确的因果边。这标志着其在实际应用中具有明显的潜力。",{"type":17,"tag":25,"props":288,"children":289},{},[290],{"type":17,"tag":31,"props":291,"children":292},{},[293],{"type":23,"value":294},"05",{"type":17,"tag":25,"props":296,"children":297},{},[298],{"type":17,"tag":31,"props":299,"children":300},{},[301],{"type":23,"value":302},"总结与展望",{"type":17,"tag":25,"props":304,"children":305},{},[306],{"type":23,"value":307},"本文研究了如何在观测数据上学习多粒度因果结构，这是一个实际且重要的问题，然而目前的相关研究较少。本文提出了一种有效的方法MgCSL，通过利用编码独立的稀疏自编码器提取潜在的宏观变量，并借助多层感知器对因果机制进行建模，进而引入简化无环惩罚加速多粒度因果结构的发现。在合成和真实数据集上的实验证明了MgCSL的有效性。",{"type":17,"tag":25,"props":309,"children":310},{},[311],{"type":23,"value":312},"在使用昇思MindSpore框架复现算法时，按照官方提供的文档和示例，能够较为轻松地完成大部分代码的实现，而昇思MindSpore的易用性与并行加速功能显著地提升了开发效率，为开发者带来了便利。随着越来越多的开发者与研究者加入昇思MindSpore社区，期待大家能够共同建设一个开放、协作的平台，积极分享使用经验并提出改进建议，推动昇思MindSpore的持续进步和不断完善。",{"type":17,"tag":25,"props":314,"children":315},{},[316],{"type":23,"value":317},"往期回顾",{"type":17,"tag":18,"props":319,"children":320},{"id":7},[321],{"type":17,"tag":72,"props":322,"children":325},{"href":323,"rel":324},"http://mp.weixin.qq.com/s?__biz=MzkxMTM2MjMzNg==&mid=2247614420&idx=1&sn=8d9c49e82b9b330a1b7ff0fa3a5a24ac&chksm=c11e2f9bf669a68d9f500863e9c15fc324aa8d37ddff506d6c8b71ff4bc4ac391aacfc55f076&scene=21#wechat_redirect",[76],[],{"type":17,"tag":25,"props":327,"children":328},{},[329,336,338,345,346,353,354],{"type":17,"tag":72,"props":330,"children":333},{"href":331,"rel":332},"http://mp.weixin.qq.com/s?__biz=MzkxMTM2MjMzNg==&mid=2247616069&idx=1&sn=40a89de0df0ee584097992d15f804809&chksm=c11e160af6699f1cc5530e86f404880fcbc5e306ea6ba5ea61a800de8e788cc58e4cd5e2803c&scene=21#wechat_redirect",[76],[334],{"type":23,"value":335},"论文精讲 | 基于昇思MindSpore评测的FLAG3D——自然语言引导的三维健身动作数据集",{"type":23,"value":337}," ",{"type":17,"tag":72,"props":339,"children":342},{"href":340,"rel":341},"http://mp.weixin.qq.com/s?__biz=MzkxMTM2MjMzNg==&mid=2247615411&idx=1&sn=3473cb214717cdbb546a91d3bc905756&chksm=c11e13fcf6699aeabe0978a6ff8273dc41e50adf9c34c3597a494c51b71d2eabf9aa1cd53465&scene=21#wechat_redirect",[76],[343],{"type":23,"value":344},"论文精讲 | 基于昇思MindSpore的加速对抗训练算法，可显著减少训练时间",{"type":23,"value":337},{"type":17,"tag":72,"props":347,"children":350},{"href":348,"rel":349},"http://mp.weixin.qq.com/s?__biz=MzkxMTM2MjMzNg==&mid=2247614719&idx=1&sn=2d49a081519f04194a6d16d9c880176f&chksm=c11e10b0f66999a690d5110bf5d339b133013bd2d7d9d4275159a816f4058be23b14c430b059&scene=21#wechat_redirect",[76],[351],{"type":23,"value":352},"论文精讲 | 基于昇思MindSpore的联合矫正与细化平衡低光照图像增强",{"type":23,"value":337},{"type":17,"tag":72,"props":355,"children":357},{"href":323,"rel":356},[76],[358],{"type":23,"value":359},"论文精讲 | 基于昇思MindSpore片段级异常注意力的弱监督视频异常检测",{"title":7,"searchDepth":361,"depth":361,"links":362},4,[],"markdown","content:technology-blogs:zh:3099.md","content","technology-blogs/zh/3099.md","technology-blogs/zh/3099","md",1776506125949]