{"pdfData":[[{"bbox":[86,79,295,92],"type":"text","angle":0,"index":0,"text":"图-解决模式识别（Pattern Recognize 问题）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"5e31f506-34bb-48b8-a573-91075ab54887","page_size":[595,841],"block_position":"0-0"},{"bbox":[87,95,224,106],"type":"text","angle":0,"index":1,"text":"Node occurance 边的关联性","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"89f3a793-0a89-4eb8-a35f-a75f5e17f4a3","page_size":[595,841],"block_position":"0-1"},{"bbox":[87,110,142,121],"type":"text","angle":0,"index":2,"text":"节点的属性","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"72d877ee-67d1-4dd3-af9a-1e02db527925","page_size":[595,841],"block_position":"0-2"},{"bbox":[87,126,132,137],"type":"text","angle":0,"index":3,"text":"边的信息","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"8eaf2ebd-8125-469e-8fd6-60a381b7cd56","page_size":[595,841],"block_position":"0-3"},{"bbox":[90,139,144,153],"type":"interline_equation","angle":0,"index":4,"text":"$$\nX _ {V} \\in \\mathbb {R} ^ {| V | \\times d}\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":0,"id":"9f59e3a4-3da2-4cb1-b0c7-5fa6d4289bcc","page_size":[595,841],"block_position":"0-4"},{"bbox":[89,158,146,172],"type":"interline_equation","angle":0,"index":5,"text":"$$\nX _ {E} \\in \\mathbb {R} ^ {| E | \\times d _ {E}}\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":0,"id":"bafdd14d-ee3d-43bf-999c-aae85401c473","page_size":[595,841],"block_position":"0-5"},{"bbox":[86,188,508,247],"type":"text","angle":0,"index":6,"text":"动态图 在紧凑的表示中集成拓扑和时间信息能力，更多用于建模动态系统，如社交网络预测、推荐系统、交通预测、谣言传播等等，结合了顺序/时序数据处理和静态图学习的挑战。图被要求处理随时间变化的动态的拓扑结构（and/or）（边数）属性（节点属性边属性），为了对动态系统建模。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"e9fcc190-da18-4caa-8943-53b87a142569","page_size":[595,841],"block_position":"0-6"},{"bbox":[87,250,357,263],"type":"text","angle":0,"index":7,"text":"对顺序数据进行处理，捕获序列不同实体之间的依赖关系。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"041a259e-5a84-4710-af6d-8b0f8a72c6f0","page_size":[595,841],"block_position":"0-7"},{"bbox":[87,265,240,278],"type":"text","angle":0,"index":8,"text":"主要有 LSTM 的卷积架构所取代。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"bab8effb-9c42-4d6c-85b8-8440d68a8eed","page_size":[595,841],"block_position":"0-8"},{"bbox":[86,281,430,294],"type":"text","angle":0,"index":9,"text":"基于解码器/编码器框架的序列模型，如 Transformer、依赖于注意力机制","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"cad79066-7fe6-45c7-87d0-458239210288","page_size":[595,841],"block_position":"0-9"},{"bbox":[86,297,444,310],"type":"text","angle":0,"index":10,"text":"在静态图学习，克服图表示中缺乏节点排序所固有的置换不变性/等变性约束。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"a1cf26dc-1d8c-4428-89d1-b7608dc08131","page_size":[595,841],"block_position":"0-10"},{"bbox":[87,312,249,325],"type":"text","angle":0,"index":11,"text":"解决方法消息传递神经网络 MPNN","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"2cd7b073-287a-407e-84b2-9f802ab82b1a","page_size":[595,841],"block_position":"0-11"},{"bbox":[87,328,139,340],"type":"text","angle":0,"index":12,"text":"不同的 DG","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"4d1258e9-9326-4fab-9cf9-78df49af7114","page_size":[595,841],"block_position":"0-12"},{"bbox":[87,343,157,356],"type":"text","angle":0,"index":13,"text":"离散的/连续的","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"11d46710-a6ca-43fa-8273-ded0b2c3d2d1","page_size":[595,841],"block_position":"0-13"},{"bbox":[87,359,213,371],"type":"text","angle":0,"index":14,"text":"边进化/节点进化/属性进化","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"e316cec7-ebd0-4678-8ecf-d424ee532c5f","page_size":[595,841],"block_position":"0-14"},{"bbox":[87,375,142,387],"type":"text","angle":0,"index":15,"text":"同质与异质","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"aeba1f87-955b-4d40-90b2-fbb527703974","page_size":[595,841],"block_position":"0-15"},{"bbox":[87,391,227,402],"type":"text","angle":0,"index":16,"text":"学习环境：直推式还是归纳式","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"c099afe5-67eb-486b-8e30-b04c0de0855f","page_size":[595,841],"block_position":"0-16"},{"bbox":[87,406,195,418],"type":"text","angle":0,"index":17,"text":"推断（前向预测）阶段","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"0c8517b0-6bb7-4bbe-90a7-98a8e6051b64","page_size":[595,841],"block_position":"0-17"},{"bbox":[87,422,171,434],"type":"text","angle":0,"index":18,"text":"通过输入X预测Y","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"b9e1a1c5-ee8a-487c-b9e7-85ead2456d73","page_size":[595,841],"block_position":"0-18"},{"bbox":[87,437,188,449],"type":"text","angle":0,"index":19,"text":"学习阶段（反向优化）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"d45a8fda-b1a8-4c3c-85b4-a07a0b5b9d2e","page_size":[595,841],"block_position":"0-19"},{"bbox":[87,453,194,465],"type":"text","angle":0,"index":20,"text":"根据Y和Y‘，优化参数","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"ae387451-65d1-409c-9cc4-0bf9850e266d","page_size":[595,841],"block_position":"0-20"},{"bbox":[86,468,504,497],"type":"text","angle":0,"index":21,"text":"如前所述，监督学习旨在学习输入空间 X和输出空间Y之间的映射函数。X可以表示的空间。本小节现在讨论空间Y。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"02ebc789-6328-424c-b5d8-7c84538febea","page_size":[595,841],"block_position":"0-21"},{"bbox":[87,499,495,513],"type":"text","angle":0,"index":22,"text":"由于动态图涉及来自图和时间/顺序数据的概念，因此必须考虑这两个方面来定义空间Y。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"fa6475a7-faa6-4c50-940b-b8f0ca1b6e48","page_size":[595,841],"block_position":"0-22"},{"bbox":[87,515,504,545],"type":"text","angle":0,"index":23,"text":"当学习序列数据的统计模型时，输入 X 可以由 T 个时间步长的 d 维特征表示，表示为X∈RTxd $\\boldsymbol { X } \\in \\mathbb { R } ^ { T \\times d }$。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"4e6dd99f-2bdf-4476-a577-5f654671189a","page_size":[595,841],"block_position":"0-23"},{"bbox":[86,561,504,622],"type":"text","angle":0,"index":24,"text":"输出Y的粒度可以是时间步长级别（每个时间步长一个输出，如在词性标记或更一般地序列标记中），或聚集的（一个输出用于许多输入，如在情感分类或更一般地序列分类中）。在静态图学习中，除了特征XV和XE之外，输入还具有拓扑信息。至于序列，输出 Y可以是局部的（每个节点或每条边一个标签）或全局的（每个子图或整个图一个标签）。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"6cf168bf-ad07-46cc-91a2-85cb8da4ddb5","page_size":[595,841],"block_position":"0-24"},{"bbox":[91,623,252,664],"index":25,"angle":0,"type":"image","img_path":"/296469f5d0a2010eb375ed7c66eed44792ebbceb0c9c733fa650c9c61c177c37.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.27058823529411763,"page_idx":0,"id":"5de26423-c844-49bd-b934-b4ebc1e7cc73","page_size":[595,841],"block_position":"0-25"},{"bbox":[129,672,188,686],"index":26,"angle":0,"type":"text","text":"Inference","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"0ba25fec-3585-4cc8-b16d-488b259ce76d","page_size":[595,841],"block_position":"0-26"},{"bbox":[312,624,421,638],"type":"text","angle":0,"index":27,"text":"Parameter Updates","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"e3ff58b0-b609-419d-ad5a-e074e7c48f1d","page_size":[595,841],"block_position":"0-27"},{"bbox":[276,639,456,666],"type":"interline_equation","angle":0,"index":28,"text":"$$\n\\mathbf {\\Theta} ^ {\\prime} = \\operatorname {u p d a t e} (\\mathbf {\\Theta}, \\hat {\\mathbf {Y}}, \\mathbf {Y})\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":0,"id":"2d0acf65-715a-45b6-a1b9-462b8f540c21","page_size":[595,841],"block_position":"0-28"},{"bbox":[86,702,504,730],"type":"text","angle":0,"index":30,"text":"许多最新的模型遵循编码器/解码器原理，其中可变长度的输入信号被编码成潜在表示，然后由解码器使用该潜在表示来计算下游任务的输出信号","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"4b041386-da4c-4422-a146-a2f04a23759d","page_size":[595,841],"block_position":"0-29"},{"bbox":[87,734,211,747],"type":"text","angle":0,"index":31,"text":"这里的 Z 就是 embedding","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"c75f970d-a01e-454b-a5c3-df741d2995b6","page_size":[595,841],"block_position":"0-30"},{"bbox":[87,749,267,762],"type":"text","angle":0,"index":32,"text":"再把 embedding 通过解码器得到预测","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":0,"id":"9ff6a07b-17f4-475a-ad00-e3d8702e3412","page_size":[595,841],"block_position":"0-31"}],[{"bbox":[88,77,503,147],"index":0,"angle":0,"type":"image","img_path":"/e2949c5b136177889bb7cc594ee5765d36bef4e6c8b7e90fe074c101bfb8a6ea.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.6974789915966386,"page_idx":1,"id":"09070cd9-e8e2-4e47-a83c-0e3f25dfdbff","page_size":[595,841],"block_position":"1-0"},{"bbox":[86,156,294,169],"type":"text","angle":0,"index":1,"text":"输入和输出序列的单元可能具有不同的顺序。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"8e2892f8-2c36-44b6-bbb4-ff5a04ad0a1f","page_size":[595,841],"block_position":"1-1"},{"bbox":[86,172,304,185],"type":"text","angle":0,"index":2,"text":"学习潜在表示（也称为嵌入）被称为表示学习。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"1e09d861-5422-413c-b376-ce312f5425e4","page_size":[595,841],"block_position":"1-2"},{"bbox":[86,188,357,201],"type":"text","angle":0,"index":3,"text":"当有多个节点/边的类型（项目和用户），用类型映射函数","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"3787ce1a-fcec-4f1d-8e11-340f8c7f4530","page_size":[595,841],"block_position":"1-3"},{"bbox":[86,218,175,230],"type":"text","angle":0,"index":4,"text":"GNN/GAT/K-GNN","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"58f599de-0c0e-4795-a586-af2713561826","page_size":[595,841],"block_position":"1-4"},{"bbox":[87,234,202,247],"type":"text","angle":0,"index":5,"text":"DG 动态图——两种形式","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"1e0c2a83-127b-44f0-95c0-70ca97149862","page_size":[595,841],"block_position":"1-5"},{"bbox":[87,250,164,263],"type":"text","angle":0,"index":6,"text":"一种离散形式的","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"d2cfee3b-e5eb-487e-ace1-ae16f0ba49c2","page_size":[595,841],"block_position":"1-6"},{"bbox":[87,266,119,277],"type":"text","angle":0,"index":7,"text":"DTDG","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"98f9ff67-2b77-4aa9-a58b-d6cc7caa34e8","page_size":[595,841],"block_position":"1-7"},{"bbox":[87,281,302,295],"type":"text","angle":0,"index":8,"text":"只要记住每一个时间段的快照（snapshot->ti）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"d3dc34d1-ecba-4427-8312-d02770847762","page_size":[595,841],"block_position":"1-8"},{"bbox":[87,297,192,310],"type":"text","angle":0,"index":9,"text":"DTDG $=$ （G1,G2…,GT）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"261f4d26-d538-4198-9df5-1de2039f7bb8","page_size":[595,841],"block_position":"1-9"},{"bbox":[87,312,164,325],"type":"text","angle":0,"index":10,"text":"一种连续形式的","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"8b7d276d-1c34-4455-9a0e-c22b928a0ef1","page_size":[595,841],"block_position":"1-10"},{"bbox":[87,328,193,340],"type":"text","angle":0,"index":11,"text":"CTDG 连续时间动态图","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"ed76639f-1466-4471-837c-e3928019b7fa","page_size":[595,841],"block_position":"1-11"},{"bbox":[87,343,175,356],"type":"text","angle":0,"index":12,"text":"基于事件流的形式","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"8a84324d-514b-4480-89a8-a34f4748c4bd","page_size":[595,841],"block_position":"1-12"},{"bbox":[87,359,216,372],"type":"text","angle":0,"index":13,"text":"每一次事件的变化记录下来","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"99f77cc6-b19d-480c-8f23-37f3ae84082f","page_size":[595,841],"block_position":"1-13"},{"bbox":[87,375,265,387],"type":"text","angle":0,"index":14,"text":"瞬时交互两个节点的（u，v）和时间 t","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"d06a53f6-a6b7-4288-a2fd-8950a541cef3","page_size":[595,841],"block_position":"1-14"},{"bbox":[87,391,235,404],"type":"text","angle":0,"index":15,"text":"Contact -Sequence ={(ui,vi,ti)}","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"e4fa3f3f-0927-4562-8488-b0b8bc039bc4","page_size":[595,841],"block_position":"1-15"},{"bbox":[86,406,504,434],"type":"text","angle":0,"index":16,"text":"Event-Based={(ui,vi,ti,deltai)} 起始(出现 )时间 ti 和持续时间 delta i，两个节点关联的边的存在的时刻","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"f236bfcf-dfe4-4d12-a8d2-4f301e69fce8","page_size":[595,841],"block_position":"1-16"},{"bbox":[86,438,437,450],"type":"text","angle":0,"index":17,"text":"Interval-Graph={(ui,vi,Te);Te=((t1,t1’),(t2,t2’))}，用 Te 来表示边的所有活动","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"d270d777-c9e1-4e65-8063-f61a82e91214","page_size":[595,841],"block_position":"1-17"},{"bbox":[86,453,505,481],"type":"text","angle":0,"index":18,"text":"Graph stream 大规模图，用于边的添加和减少 delta i=1 add or -1deletion 标志位不考虑持续时间","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"a937f0ac-3510-4144-9d16-ff345775196b","page_size":[595,841],"block_position":"1-18"},{"bbox":[86,484,279,497],"type":"text","angle":0,"index":19,"text":"Graph Stream={（ui，vi，ti，delta i）}","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"eb2e67e8-3481-4052-a808-a0ce59780579","page_size":[595,841],"block_position":"1-19"},{"bbox":[87,499,138,512],"type":"text","angle":0,"index":20,"text":"什么事件？","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"ccef197d-be95-40e6-93d6-ba42460f5a7b","page_size":[595,841],"block_position":"1-20"},{"bbox":[86,515,264,528],"type":"text","angle":0,"index":21,"text":"Ti时刻下，ui和vi的边增加了/减少了","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"264a7380-a403-4eaf-828a-40ae3fd6c416","page_size":[595,841],"block_position":"1-21"},{"bbox":[126,534,183,544],"index":22,"angle":0,"type":"image_caption","text":"Static Graphs","id":"d11c3a29-93e0-45d6-9da5-c9e44c7ccc11","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-22"},{"bbox":[126,548,201,557],"index":23,"angle":0,"type":"image_caption","text":"Weighted directed graph","id":"3d4cfa3d-02ff-48b2-a352-d609bcdc82a0","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-23"},{"bbox":[132,559,188,618],"index":24,"angle":0,"type":"image","img_path":"/c746830d302e3e44284fd48b73c0af69a11d0acf4ee6059f4189b80414d69c7f.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":1,"id":"7bec619a-004d-4c3e-a280-41e2247ef7c2","page_size":[595,841],"block_position":"1-24"},{"bbox":[126,624,190,633],"index":25,"angle":0,"type":"image_caption","text":"Heterogeneous graph","id":"ae3c366b-01b6-45a5-a01d-824c0fb712de","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-25"},{"bbox":[133,635,188,694],"index":26,"angle":0,"type":"image","img_path":"/6f1bb89499913ea46bf89c5766e785d288ad81eb6d1d4371993b350c787bd28d.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":1,"id":"52e0bde5-081f-4d82-b376-a0d2811090d7","page_size":[595,841],"block_position":"1-26"},{"bbox":[220,534,290,545],"index":27,"angle":0,"type":"image_caption","text":"Dynamic Graphs","id":"0c110420-5c9a-4211-bc4f-9237e09ae55a","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-27"},{"bbox":[220,548,302,557],"index":28,"angle":0,"type":"image_caption","text":"DTDG (or Snapshot vision)","id":"91d60587-7245-4e9a-96b3-b1bb4b79fbd4","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-28"},{"bbox":[220,560,348,615],"index":29,"angle":0,"type":"image","img_path":"/3e4e316277ba51154368e0a57961bebe492938e98fcdc466f4510639ac5629e6.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.21512605042016808,"page_idx":1,"id":"5f0c7577-6fee-4056-85b7-3e9facf506e9","page_size":[595,841],"block_position":"1-29"},{"bbox":[221,624,293,633],"index":30,"angle":0,"type":"image_caption","text":"CTDG (or Event vision)","id":"c80e3cd4-7ac1-4dca-ba2e-a713ca169182","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-30"},{"bbox":[220,637,348,694],"index":31,"angle":0,"type":"image","img_path":"/8ef90f4621dfc41510d57405f3c6bb283fb266f06ec85835cad4eb31d5ae5683.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.21512605042016808,"page_idx":1,"id":"9cf9a3bb-95cd-4dd3-8bec-fe8e77e943fa","page_size":[595,841],"block_position":"1-31"},{"bbox":[359,548,421,557],"index":32,"angle":0,"type":"image_caption","text":"(Edge-oriented) ESG","id":"57855d91-e8ba-4015-9515-bd853684c656","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-32"},{"bbox":[358,571,527,608],"index":33,"angle":0,"type":"image","img_path":"/0681e7b4f29f3c5eba8a879c82bd10665da1edd827bdf1c903ae625ea3f30b1c.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.28403361344537814,"page_idx":1,"id":"5d414a3d-e628-4054-9622-d386dc52e977","page_size":[595,841],"block_position":"1-33"},{"bbox":[358,624,422,633],"index":34,"angle":0,"type":"image_caption","text":"(Node-oriented) ESG","id":"43a3dbae-0a53-4210-8ccf-36b024a49fb6","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-34"},{"bbox":[358,646,521,682],"index":35,"angle":0,"type":"image","img_path":"/c758284426fe875cbb2097ef354f0d4a80e757498dc69b3b2a64c06ba310b2a0.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2739495798319328,"page_idx":1,"id":"3b6c52f3-c744-4970-8c79-355c4d0a465f","page_size":[595,841],"block_position":"1-35"},{"bbox":[86,717,274,730],"type":"text","angle":0,"index":36,"text":"如何将动态图转换成等价的静态图ESG？","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"id":"de82167b-eefc-49ed-8882-0f598fb82ae1","page_size":[595,841],"block_position":"1-36"},{"bbox":[86,733,483,746],"type":"text","angle":0,"index":37,"text":"1、基于边的形式 节点的备份 $^ +$ 边的联系（节点间 $^ +$ 时间戳，加入了节点之间的关联性）","id":"0663d560-2bc3-43a2-99c2-3f00823a67c1","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-37"},{"bbox":[87,749,182,761],"type":"text","angle":0,"index":38,"text":"2、基于节点的形式","id":"37ce1305-b19c-4ea2-9a0a-e0bf5d4f8bcf","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":1,"page_size":[595,841],"block_position":"1-38"}],[{"bbox":[105,79,260,91],"type":"text","angle":0,"index":0,"text":"GNNDelete如何删除边的关联性","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"ea8e41fd-b11c-4180-9740-27ab6ef81621","page_size":[595,841],"block_position":"2-0"},{"bbox":[86,95,147,106],"type":"title","angle":0,"index":1,"text":"# 定义动态度：","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"254ea28c-012b-4776-9225-1d9328f115e6","page_size":[595,841],"block_position":"2-1"},{"bbox":[87,110,504,138],"type":"text","angle":0,"index":2,"text":"1、边和顶点都固定，fix V,E 时空图（STGCN 时空模型，既要学时间上的， 又要学空间上）","id":"2325d4dc-edfd-41de-bb5b-91bc93146673","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-2"},{"bbox":[87,141,194,153],"type":"text","angle":0,"index":3,"text":"2、只固定了顶点,fix V","id":"1d6c5c8b-6888-489c-a931-08649e153e73","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-3"},{"bbox":[87,157,171,170],"type":"text","angle":0,"index":4,"text":"3、都变化，vary","id":"7e141ba6-110e-49d4-9e55-f745eca720b8","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-4"},{"bbox":[87,172,264,185],"type":"text","angle":0,"index":5,"text":"4、边集不变，节点集变化，没有意义","id":"3b76ebe3-1f60-485f-b192-8f4464a4cbb5","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-5"},{"bbox":[87,203,158,216],"type":"title","angle":0,"index":7,"text":"# 输出是有粒度：","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"ac2c91a5-7ca1-439e-bb9a-1b2e1227ecec","page_size":[595,841],"block_position":"2-6"},{"bbox":[87,219,185,232],"type":"title","angle":0,"index":8,"text":"# 预测单一时刻的结果","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"36d4ebee-52a8-43b7-9fae-c88244ce2a02","page_size":[595,841],"block_position":"2-7"},{"bbox":[87,235,164,247],"type":"title","angle":0,"index":9,"text":"# 还是多个时刻的","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"cf21bbb4-80de-44cc-8577-f14f1dbc7dd9","page_size":[595,841],"block_position":"2-8"},{"bbox":[88,250,296,263],"type":"text","angle":0,"index":10,"text":"：时间步长级别的（ 每个时间步长一个输出）","id":"1694b175-341d-4253-adfc-8c2f6eac119c","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-9"},{"bbox":[88,266,356,279],"type":"text","angle":0,"index":11,"text":"：聚合的，一个输出多个输入（未来五个时间的拓扑结构）","id":"89b25048-a50b-4988-8fae-fd25fd16ca99","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"page_size":[595,841],"block_position":"2-10"},{"bbox":[87,281,258,294],"type":"text","angle":0,"index":13,"text":"拓扑结构也有局部（子图）和全局的","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"32102220-93a2-4be5-97cb-26d4a56a8eaa","page_size":[595,841],"block_position":"2-11"},{"bbox":[87,296,451,325],"type":"text","angle":0,"index":14,"text":"直推式 原来的拓扑结构不变 做出预测的时候没有在训练过程中没出现过的节点归纳式 做出推断的时候训练过程中没有看到","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"ced84715-c2a7-4bec-a9e2-157cafbd9c48","page_size":[595,841],"block_position":"2-12"},{"bbox":[91,328,214,348],"type":"title","angle":0,"index":15,"text":"# Transductive","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"3056cd25-572b-45fc-933f-677ab20a135b","page_size":[595,841],"block_position":"2-13"},{"bbox":[91,353,230,386],"type":"text","angle":0,"index":16,"text":"(1) Transductive-fixv,ELearning Inference","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"06dd9d6c-8b14-4dcd-8ca6-165f7892f38b","page_size":[595,841],"block_position":"2-14"},{"bbox":[93,386,157,445],"index":17,"angle":0,"type":"image","img_path":"/bbcc93ac856be6d7c461092f4c7067219dd077d1ab5b3a264c057c10808b45aa.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"18502830-4eaf-4dff-a3b0-fdfa1fdef1ff","page_size":[595,841],"block_position":"2-15"},{"bbox":[160,386,227,445],"index":18,"angle":0,"type":"image","img_path":"/db4d98da46d1e6c6e158c05a5859fb732876cbad79b0288d64c35862f9805c4c.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"c01e58d3-9bba-472c-80eb-b738261ade71","page_size":[595,841],"block_position":"2-16"},{"bbox":[248,353,380,386],"type":"text","angle":0,"index":19,"text":"(2) Transductive-fixyLearning Inference","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"15ef4726-0f6c-47e5-8d14-b2aa8a4087ef","page_size":[595,841],"block_position":"2-17"},{"bbox":[249,386,312,445],"index":20,"angle":0,"type":"image","img_path":"/5d77a9975b12967c827a904f0e653a6dd20b616eb17170ea86c227b4bff8c8af.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"291b3f94-1396-4cce-8cc4-0fb41caeb7db","page_size":[595,841],"block_position":"2-18"},{"bbox":[318,386,384,445],"index":21,"angle":0,"type":"image","img_path":"/fae214f407c72c98ec8787121c73e96333ea90b02082385935af350087cea3a4.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"0346e4d1-d58a-4456-b9c4-d28ad9cf7b86","page_size":[595,841],"block_position":"2-19"},{"bbox":[397,353,530,386],"type":"text","angle":0,"index":22,"text":"(3) Transductive-varyLearning Inference","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"03fd2912-5fb6-4754-a198-53ab55a6d6d4","page_size":[595,841],"block_position":"2-20"},{"bbox":[399,386,462,445],"index":23,"angle":0,"type":"image","img_path":"/e69e65651cde1ce432bc78ee711dfe7f2ab620c85a18ec5abadd2b7dc03c09ce.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"8c62edef-17b0-4489-9e90-eb625cd6fec1","page_size":[595,841],"block_position":"2-21"},{"bbox":[467,386,533,445],"index":24,"angle":0,"type":"image","img_path":"/d3f2b68951d63e00609989797d595e11f69387b3ae4a9d49ddf6e70f76b236b8.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"4c6095c1-c433-48a7-b016-2c75a36c934d","page_size":[595,841],"block_position":"2-22"},{"bbox":[91,459,181,478],"type":"title","angle":0,"index":25,"text":"# Inductive","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":2,"id":"a586546e-ed02-495a-9fc5-db6df367f51b","page_size":[595,841],"block_position":"2-23"},{"bbox":[91,481,196,499],"type":"text","angle":0,"index":26,"text":"（4） Inductivev","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"af017829-45fa-4cec-bd89-5b2627f86279","page_size":[595,841],"block_position":"2-24"},{"bbox":[93,500,238,573],"index":27,"angle":0,"type":"image","img_path":"/37efc40a1a19ed74b9949a8171bf9ac0dee51adfcf523d66b3d6c78e6feb4573.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.24369747899159663,"page_idx":2,"id":"05ba68d3-1e04-43e4-95ea-d93f8d0717b4","page_size":[595,841],"block_position":"2-25"},{"bbox":[248,481,361,500],"type":"text","angle":0,"index":28,"text":"(5) InductiveDG","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"5b33407b-1a88-4c1a-9a12-b1d66d7eaadd","page_size":[595,841],"block_position":"2-26"},{"bbox":[249,500,331,572],"index":29,"angle":0,"type":"image","img_path":"/cf63288df9720a4b9ad1e8a1ad001c61714dd67c023f799376cc80871a30dffe.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"da7e91ce-b0dc-417b-a769-72ef054ae0d6","page_size":[595,841],"block_position":"2-27"},{"bbox":[333,500,414,573],"index":30,"angle":0,"type":"image","img_path":"/8df199df3a9c6c47dd44f0eda1352bfe46da210844d723b3f9bda3fbbf91d868.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":2,"id":"b8927f6c-64e5-4572-8e7a-c607099205bf","page_size":[595,841],"block_position":"2-28"},{"bbox":[86,577,509,606],"type":"text","angle":0,"index":31,"text":"离散的时间直推式学习中（DTDG,Transductive），transfix VE ：交通流量，传染病病例，犯罪数量，作物产量","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"77a4e668-f2bb-4ee8-8db0-18c573d25ff7","page_size":[595,841],"block_position":"2-29"},{"bbox":[86,608,477,636],"type":"text","angle":0,"index":32,"text":"图/全局任务：每个快照的分类 睡眠阶段分类；整个DG的预测，如骨骼的动作STG；Tarnsfix V：电信网络中连通性，会议中个人联系","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"2a233e02-b2ca-4bc6-b74b-ac6e09ff23ac","page_size":[595,841],"block_position":"2-30"},{"bbox":[86,640,473,667],"type":"text","angle":0,"index":33,"text":"离散的归纳式学习（DTDG Inductive）中 社交网络未来快照的节点分类或链接预测图集输出 基于社交网络传播树的快照对真是和假新闻分类","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"296923b9-5e91-4898-9d74-fddd9c8e4ade","page_size":[595,841],"block_position":"2-31"},{"bbox":[87,671,174,683],"type":"text","angle":0,"index":34,"text":"连续时间（CTDG）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"ea7d524b-9f7b-42e2-9962-867659e5e87b","page_size":[595,841],"block_position":"2-32"},{"bbox":[87,687,185,699],"type":"text","angle":0,"index":35,"text":"推荐系统、社交网络","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"fb4356fd-d8e7-4731-8d44-b67b1356e3fa","page_size":[595,841],"block_position":"2-33"},{"bbox":[86,702,478,745],"type":"text","angle":0,"index":36,"text":"CTCG再其最小时间单元下不能访问全局/整个图信息，全局的时间步标签是无意义的但是全局聚合任务可以通过聚合不同时间步长的节点来实现，如连续时间内谣言检测CTDG也可以定期拍摄快照转换为DTDG","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"2d46b937-35dd-485e-82a4-ce0f8ddd0770","page_size":[595,841],"block_position":"2-34"},{"bbox":[87,749,173,761],"type":"text","angle":0,"index":37,"text":"动态时间下的任务","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":2,"id":"50888c31-eab4-496c-af87-1af9af0124d6","page_size":[595,841],"block_position":"2-35"}],[{"bbox":[86,79,342,92],"type":"text","angle":0,"index":0,"text":"为了评估给定任务的统计模型的性能，传统机器学学习","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"b93d59a0-6523-4e70-a70c-2d2772d6620f","page_size":[595,841],"block_position":"3-0"},{"bbox":[87,95,132,107],"type":"text","angle":0,"index":1,"text":"分类任务","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"52858037-29a2-40e5-90a5-205786bc7ebf","page_size":[595,841],"block_position":"3-1"},{"bbox":[87,110,132,122],"type":"text","angle":0,"index":2,"text":"回归任务","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"4c8e459d-1452-4636-ac22-4e5373fb5762","page_size":[595,841],"block_position":"3-2"},{"bbox":[87,125,152,138],"type":"text","angle":0,"index":3,"text":"节点排名任务","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"c9c62654-c389-46fa-9541-c7fc35c55f1c","page_size":[595,841],"block_position":"3-3"},{"bbox":[86,141,249,154],"type":"text","angle":0,"index":4,"text":"动态任务 沿着时间轴计算静态度量","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"ef6d1da2-4711-455f-a683-ae0db739dcea","page_size":[595,841],"block_position":"3-4"},{"bbox":[87,157,152,169],"type":"text","angle":0,"index":5,"text":"动态图的嵌入","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"c02e48cd-5939-4fa6-b72b-2f22b1939b48","page_size":[595,841],"block_position":"3-5"},{"bbox":[86,171,509,249],"type":"text","angle":0,"index":6,"text":"从编码器-解码器的角度来看，深度学习统计模型首先将原始输入映射到表示为 Z的嵌入中，然后利用 Z 来预测输出[12，4]。图形可以在节点/边缘级别或（子）图形级别嵌入[13，5].节点级嵌入有利于广泛的节点相关任务，并允许保留更完整的输入信息以供稍后计算[5]以同样的方式，时间步长级嵌入比时间聚合嵌入保留更多的信息。类似地，当在顺序数据上学 习 时 [7] 。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"a65ea61c-d467-4376-96fd-dfc1fc8c3115","page_size":[595,841],"block_position":"3-6"},{"bbox":[91,255,490,520],"index":7,"angle":0,"type":"image","img_path":"/c3a808e669a8b7be887e2835a878309fb07a5f2690536f0e5cc4d52868f7a209.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.6705882352941176,"page_idx":3,"id":"04e9b88d-1bd8-4257-80e8-258636e45436","page_size":[595,841],"block_position":"3-7"},{"bbox":[88,528,190,544],"type":"interline_equation","angle":0,"index":8,"text":"$$\nZ \\in \\mathbb {R} ^ {| V | \\times | T | \\times d}, z _ {v} ^ {t} \\in \\mathbb {R} ^ {d}\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":3,"id":"6f748d9a-4464-4462-a0f4-cdfefb3f447d","page_size":[595,841],"block_position":"3-8"},{"bbox":[86,561,317,577],"type":"text","angle":0,"index":9,"text":"对不同的动态图， 如何设计 encoder 的 settings","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":3,"id":"8ff6572d-f913-451a-8139-9ebac7dd3b3d","page_size":[595,841],"block_position":"3-9"}],[{"bbox":[107,78,271,93],"index":0,"angle":0,"type":"image_caption","text":"(A) Discrete, transductive","id":"c4973036-2a31-4212-b421-6c103a96a8de","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-0"},{"bbox":[91,94,283,245],"index":1,"angle":0,"type":"image","img_path":"/46cb1d54b5358d567f25b2593621409ab8028885478293acb70c4cf8f849ab88.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.3226890756302521,"page_idx":4,"id":"f5df8d5b-3f57-4643-a0f4-917e9a19bf06","page_size":[595,841],"block_position":"4-1"},{"bbox":[337,78,480,93],"index":2,"angle":0,"type":"image_caption","text":"(B) Discrete, inductive","id":"e7402059-5fdc-4679-b3e5-8a9b384ca3e7","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-2"},{"bbox":[313,94,493,243],"index":3,"angle":0,"type":"image","img_path":"/d2df25f6574b3cf24a96ef20688baf948c6322632b4acd9e8e97597ec9a56188.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.3025210084033613,"page_idx":4,"id":"c5652455-87f8-4181-abe2-e6a2c32f155f","page_size":[595,841],"block_position":"4-3"},{"bbox":[95,264,282,280],"index":4,"angle":0,"type":"image_caption","text":"(C) Continuous, transductive","id":"7ad984d6-52ef-4d6a-9a4d-68ce0e5f5732","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-4"},{"bbox":[91,282,282,412],"index":5,"angle":0,"type":"image","img_path":"/92b7d0fd700043ee860b01b2e882046208c24d7d786f335fd8b74b18954e36f3.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.32100840336134456,"page_idx":4,"id":"1eb6faa2-41e3-4ee3-92b7-9c37bd8593cc","page_size":[595,841],"block_position":"4-5"},{"bbox":[326,264,491,280],"index":6,"angle":0,"type":"image_caption","text":"(D) Continuous, inductive","id":"3244c416-6f70-4316-9ad8-77dc8117d0ee","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-6"},{"bbox":[314,282,493,407],"index":7,"angle":0,"type":"image","img_path":"/731d08cee749b45a4111054e790163f2db076ce12de24586a28e861c309eb247.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.30084033613445377,"page_idx":4,"id":"b1d49966-e793-4f98-a13b-90c5ac28b2cf","page_size":[595,841],"block_position":"4-7"},{"bbox":[86,436,506,483],"type":"text","angle":0,"index":8,"text":"为了将动态图信息化地编码为张量或向量列表，DGNN必须捕获结构信息及其随时间的演化 。 因 此 ， 为 了 分 别 处 理 拓 扑 和 时 间 ，DG 通常被 分 解 或转换 为 等效静 态 （ 子 ） 图[74，89，75]，随机游走[113，124，123，121，122]，或者是矩阵序列[95，86，93]。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"id":"2c5a03fa-59a7-4539-bebb-4bbd05dd39f9","page_size":[595,841],"block_position":"4-8"},{"bbox":[86,514,505,560],"type":"text","angle":0,"index":9,"text":"在文献中，通过将不同的静态图编码器 fG（·）与时态数据的fT（·）相结合，出现了许多方法。大量的图和时态数据编码器是本节中回顾的 DG编码器的基础。这些编码器在附录B和C中进行了描述。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"id":"743d3a60-1db8-47cf-9561-a8ac89dba8a1","page_size":[595,841],"block_position":"4-9"},{"bbox":[86,592,499,622],"type":"text","angle":0,"index":10,"text":"在下面的小节中，我们提出了一个DGNN模型的分类，它依赖于五个类别。我们的分类，如图7所示，是基于处理时间和结构信息的策略。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"id":"3de204d7-183c-4229-99b6-ce80002f093b","page_size":[595,841],"block_position":"4-10"},{"bbox":[87,624,444,638],"type":"text","angle":0,"index":11,"text":"1.通过拓扑对时间边缘进行建模并对 ESG 进行编码，表示为 TE（第 3.1 节）。","id":"2ce077d7-4d83-4955-891c-eff94f4c085b","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-11"},{"bbox":[87,640,359,652],"type":"text","angle":0,"index":12,"text":"2.对隐藏状态进行顺序编码，记为 enc（H）（第 3.2 节）。","id":"b6bcf750-d0ae-4c8f-a772-211e1b6a56f1","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-12"},{"bbox":[86,655,385,668],"type":"text","angle":0,"index":13,"text":"3.对 DGNN 参数进行顺序编码，表示为 enc（Θ）（第 3.3 节）。","id":"a1044c44-5d01-4457-9de8-a5db24b56852","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-13"},{"bbox":[86,671,465,684],"type":"text","angle":0,"index":14,"text":"4.通过属性嵌入发生时间t作为ESG的边缘特征，表示为emb（t）（第3.4节）。","id":"be8083aa-c1a0-4704-81e9-2fe77f8bd2aa","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-14"},{"bbox":[87,687,317,700],"type":"text","angle":0,"index":15,"text":"5.抽样因果游走，表示为 CauseRW（第 3.5 节）。","id":"f099467f-0bbf-4309-8c59-30815db0eb7a","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":4,"page_size":[595,841],"block_position":"4-15"}],[{"bbox":[135,78,176,95],"type":"text","angle":0,"index":0,"text":"1. TE","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"384eaf73-f9e5-4ed7-a704-ec3e887ddba2","page_size":[595,841],"block_position":"5-0"},{"bbox":[101,105,213,168],"index":1,"angle":0,"type":"image","img_path":"/497a6bfcb699c59d3b4a9b42e24e4270536c5f1681807f903e0e85dbc4a67ff8.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"f1235de6-a3e3-4813-8abf-f5b380912190","page_size":[595,841],"block_position":"5-1"},{"bbox":[102,177,205,194],"index":2,"angle":0,"type":"image","img_path":"/36a04b2c156c0dbfeeb1f8a5e3a2caa42895d4c195620b069cf075a54dca2717.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"8a6f988b-ca41-408a-a2eb-0cc19f7b50e3","page_size":[595,841],"block_position":"5-2"},{"bbox":[102,201,193,218],"index":3,"angle":0,"type":"image","img_path":"/84576c405221d1001ad1ab5a0f3de8a24f61ef9923085f789c09f31cc6a792f9.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"afaed60e-2c41-40d6-8a08-2fecf81306a8","page_size":[595,841],"block_position":"5-3"},{"bbox":[248,78,322,98],"type":"text","angle":0,"index":4,"text":"2. Enc(H)","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"bf01932b-1e94-43db-a493-8e2112e4d9a1","page_size":[595,841],"block_position":"5-4"},{"bbox":[231,103,345,222],"index":5,"angle":0,"type":"image","img_path":"/1bb2a53245bd5c1de66e86cdf425f6fca9fbf5d9068fee5c6037b2afd2fff0d9.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"06bbcabe-1006-42ed-879e-b8d6e80829f9","page_size":[595,841],"block_position":"5-5"},{"bbox":[379,78,452,98],"type":"text","angle":0,"index":6,"text":"3. Enc(0)","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"107e29c9-d120-41d4-a349-81c9cd5be3ca","page_size":[595,841],"block_position":"5-6"},{"bbox":[359,103,473,222],"index":7,"angle":0,"type":"image","img_path":"/11776d45b7f1d6ffcfced0128e76a2d7ea793afc930bd40b6c3b614f0c6b1320.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"6c981783-1bb3-48e2-8eb0-2c494c61832c","page_size":[595,841],"block_position":"5-7"},{"bbox":[511,78,582,98],"type":"text","angle":0,"index":8,"text":"4. Emb(t)","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"219d663e-ce0f-4640-88b7-ea20a0967d0a","page_size":[595,841],"block_position":"5-8"},{"bbox":[484,101,595,225],"index":9,"angle":0,"type":"image","img_path":"/28df9d27ec3d3faac8c88b3acad757af84feff54fee07c071e078c0dff6e42e6.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"ff72959c-d87c-49a8-b7ff-3819fba65371","page_size":[595,841],"block_position":"5-9"},{"bbox":[86,234,479,248],"type":"text","angle":0,"index":10,"text":"请注意，这五种方法并不是唯一的，也就是说，它们可以组合在一起并用于同一 DG。","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"096fc671-6e5c-4360-b1c9-6b8b5d84cf0b","page_size":[595,841],"block_position":"5-10"},{"bbox":[87,250,287,263],"type":"text","angle":0,"index":11,"text":"1、对时序的边进行建模——转换为静态图","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"0d72216a-decd-4f02-b7fd-85b24ccb62df","page_size":[595,841],"block_position":"5-11"},{"bbox":[105,265,298,280],"type":"text","angle":0,"index":12,"text":"FixV,E 考虑了时间信息（embedding ET）","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"811d9248-14c6-44dc-826e-96c938c35457","page_size":[595,841],"block_position":"5-12"},{"bbox":[138,281,210,298],"index":13,"angle":0,"type":"image_caption","text":"Snapshots","id":"1b093bfe-7b7a-4f72-9157-fc7a2145645d","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"page_size":[595,841],"block_position":"5-13"},{"bbox":[111,301,225,397],"index":14,"angle":0,"type":"image","img_path":"/22d8dcba7544ae7f4b9a780b95130b617616df0c54a80f0a285449e5d6235839.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"355adff1-4efd-4298-9bb3-737d75a9a957","page_size":[595,841],"block_position":"5-14"},{"bbox":[105,406,141,417],"type":"text","angle":0,"index":15,"text":"ASTTN","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"45e0f4f5-bc7b-4808-b754-1fecef99d6ad","page_size":[595,841],"block_position":"5-15"},{"bbox":[262,280,354,299],"index":16,"angle":0,"type":"image_caption","text":"STG with Er","id":"f82ecbe4-a420-4577-8d13-bf9f7f044551","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"page_size":[595,841],"block_position":"5-16"},{"bbox":[247,301,359,375],"index":17,"angle":0,"type":"image","img_path":"/8169e54de8b83dfaaa4b484f35da48c4abb52c1e0df89fb0f62c7e9903e8e394.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.2,"page_idx":5,"id":"f6fb60ec-b64a-4bb5-a35f-c7470f154be0","page_size":[595,841],"block_position":"5-17"},{"bbox":[398,280,517,296],"index":18,"angle":0,"type":"image_caption","text":"ST-GCN Module","id":"927d6d5f-cf43-4b50-89bf-d31535741df1","color":{"line":"rgba(13, 83, 222 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"page_size":[595,841],"block_position":"5-18"},{"bbox":[380,299,536,401],"index":19,"angle":0,"type":"image","img_path":"/6984145b6f133ae0040e209cde6ece2a90e61b23f8c1a1eedced479836a1b650.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.26218487394957984,"page_idx":5,"id":"d65b91aa-5a31-4848-95d2-d75ebf0f0ec2","page_size":[595,841],"block_position":"5-19"},{"bbox":[86,437,223,450],"type":"text","angle":0,"index":20,"text":"2、对隐藏状态进行顺序编码","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"9123542b-08eb-42fe-b517-474ef977864c","page_size":[595,841],"block_position":"5-20"},{"bbox":[105,453,265,465],"type":"text","angle":0,"index":21,"text":"对拓扑结构和时间戳信息进行编码","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"dd70b7c3-1499-4192-a83d-0a3aa936c43d","page_size":[595,841],"block_position":"5-21"},{"bbox":[105,469,126,480],"type":"text","angle":0,"index":22,"text":"RST","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"75b0f467-2c60-4d0d-93e9-c38fd83395f7","page_size":[595,841],"block_position":"5-22"},{"bbox":[106,484,141,497],"type":"text","angle":0,"index":23,"text":"DyReG","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"f96b89d7-f33a-497d-9f7f-0cebb98b7185","page_size":[595,841],"block_position":"5-23"},{"bbox":[106,500,153,513],"type":"text","angle":0,"index":24,"text":"Dyn-GCN","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"7fbacd35-7a77-4fdc-9920-2d3901515fea","page_size":[595,841],"block_position":"5-24"},{"bbox":[104,514,504,543],"type":"text","angle":0,"index":25,"text":"把通过 f(G)编码的信息经过 LSTM 或者 GRU（Sequential endocding）进行时间序列的学习","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"b5d212b1-3279-4ff2-8e4d-25d4856df932","page_size":[595,841],"block_position":"5-25"},{"bbox":[105,545,293,560],"type":"interline_equation","angle":0,"index":26,"text":"$$\nZ ^ {t} = f _ {T} \\left(\\left(f _ {G} \\left(X ^ {t}\\right)\\right)\\right), o r f _ {T} \\left(X ^ {T}\\right) \\oplus f _ {G} \\left(X ^ {t}\\right)\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":5,"id":"027afc4b-fea1-478c-839e-d0f97ad1b41a","page_size":[595,841],"block_position":"5-26"},{"bbox":[105,577,279,591],"type":"text","angle":0,"index":27,"text":"解码的时候也要用sequential去阶码","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"068ae6f1-059f-4051-bf1c-100784eb73e1","page_size":[595,841],"block_position":"5-27"},{"bbox":[86,593,357,606],"type":"text","angle":0,"index":28,"text":"3、对于新增的节点很难预测，及没有将 fG 和 fT 结合起来","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"472243d0-f65e-4dcd-8a2a-6803f26dfc9b","page_size":[595,841],"block_position":"5-28"},{"bbox":[86,608,342,622],"type":"text","angle":0,"index":29,"text":"对应的是 transformer 直接把时间戳的信息编码进去了","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"dab51f35-227f-43f7-9cd6-6356070b2421","page_size":[595,841],"block_position":"5-29"},{"bbox":[86,624,270,638],"type":"text","angle":0,"index":30,"text":"通过（fg参数）时间步来学习时间信息","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"f2923eea-249e-4b25-b530-bf77119bfcff","page_size":[595,841],"block_position":"5-30"},{"bbox":[86,640,230,653],"type":"text","angle":0,"index":31,"text":"将时间信息的编码加入到 fg 里","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"1cf12403-8699-4cc5-b549-78a8e6bc2fa3","page_size":[595,841],"block_position":"5-31"},{"bbox":[92,654,192,669],"type":"interline_equation","angle":0,"index":32,"text":"$$\n\\Theta_ {f _ {o}} ^ {t} = L S T M \\left(\\Theta^ {t - 1} _ {f _ {o}}\\right)\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":5,"id":"425b378f-4f5b-45d8-8c92-c2311a0a289b","page_size":[595,841],"block_position":"5-32"},{"bbox":[88,672,197,688],"type":"interline_equation","angle":0,"index":33,"text":"$$\n\\Theta^ {t} _ {f _ {G}} = G R U \\left(H ^ {t}, \\Theta^ {t - 1} _ {f _ {G}}\\right)\n$$","color":{"line":"rgba(230, 122, 171, 1)","fill":"rgba(230, 122, 171, 1)"},"page_idx":5,"id":"7813fd4f-19ce-4295-b2ae-c23e80579341","page_size":[595,841],"block_position":"5-33"},{"bbox":[86,702,267,715],"type":"text","angle":0,"index":34,"text":"再经过编码，就没有时间上的序列性了","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"f92ff5a0-e733-4431-946c-391c90474443","page_size":[595,841],"block_position":"5-34"},{"bbox":[87,719,100,729],"type":"text","angle":0,"index":35,"text":"4、","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"8acd6b73-a7d0-4cbd-9db6-4113669248bb","page_size":[595,841],"block_position":"5-35"},{"bbox":[86,733,289,746],"type":"text","angle":0,"index":36,"text":"还把时间作为嵌入信息，如何编码时间信息","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"e214eb5c-5a81-4320-abd2-85d26b23627e","page_size":[595,841],"block_position":"5-36"},{"bbox":[87,749,206,762],"type":"text","angle":0,"index":37,"text":"Temproal point process","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":5,"id":"4259b5c6-ae0e-46e5-a456-54bf0fe26d31","page_size":[595,841],"block_position":"5-37"}],[{"bbox":[87,79,261,93],"type":"text","angle":0,"index":0,"text":"在每个节点上加一个 t 的 embedding","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"5b180b71-eeb8-4c66-8aa2-80de2fa4ce85","page_size":[595,841],"block_position":"6-0"},{"bbox":[87,95,182,107],"type":"title","angle":0,"index":1,"text":"# 3、随机游走的方法","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":6,"id":"b44bae5d-d48a-43c2-86dd-18a658b2c3da","page_size":[595,841],"block_position":"6-1"},{"bbox":[87,110,174,122],"type":"text","angle":0,"index":2,"text":"对于异质图的方法","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"f64564b9-f594-4ced-987a-3d87a4f6f7e0","page_size":[595,841],"block_position":"6-2"},{"bbox":[87,126,179,137],"type":"text","angle":0,"index":3,"text":"怎么设计一个 GNN","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"e8ffe1be-b1d4-4454-8a9f-14d083677225","page_size":[595,841],"block_position":"6-3"},{"bbox":[87,142,160,153],"type":"text","angle":0,"index":4,"text":"设计 Workflow","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"7d3ed205-7415-4021-aacb-a4fd7faa76d7","page_size":[595,841],"block_position":"6-4"},{"bbox":[87,157,153,169],"type":"text","angle":0,"index":5,"text":"如何优化框架","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"4be94b22-bd3f-43fb-80de-3e34d34865ba","page_size":[595,841],"block_position":"6-5"},{"bbox":[87,173,163,185],"type":"title","angle":0,"index":6,"text":"# 时序图神经网络","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":6,"id":"43ca5685-041e-41d8-b333-81d6d2dd806c","page_size":[595,841],"block_position":"6-6"},{"bbox":[88,188,265,200],"type":"title","angle":0,"index":7,"text":"# 1、创建数据，根据快照的方式构建图","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"level":1,"page_idx":6,"id":"a03ceb01-e69d-41db-b7d3-2a7479792e39","page_size":[595,841],"block_position":"6-7"},{"bbox":[106,200,564,459],"index":8,"angle":0,"type":"image","img_path":"/215cfded78261b39f639e96eaae2b4b9cfafc37a05f9efefaa0359da4fcafbee.jpg","color":{"line":"rgba(89, 92, 220, 1)","fill":"rgba(89, 92, 220, 1)"},"img_ratio":0.7697478991596639,"page_idx":6,"id":"7835e6b2-9b14-4a04-9c0a-f5234d8a6131","page_size":[595,841],"block_position":"6-8"},{"bbox":[105,468,333,481],"type":"text","angle":0,"index":9,"text":"表示节点的特征，如果没有特征用 one -hot 表示","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"b05956cd-ed71-4d3a-b92f-266c8d2addca","page_size":[595,841],"block_position":"6-9"},{"bbox":[106,484,199,497],"type":"text","angle":0,"index":10,"text":"计算 attention 系数","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"f0c1e921-3126-48fa-bc3e-098384f15f1a","page_size":[595,841],"block_position":"6-10"},{"bbox":[106,500,160,513],"type":"text","angle":0,"index":11,"text":"时间注意力","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"2270129d-461b-4e26-95f4-db20e62e3a3a","page_size":[595,841],"block_position":"6-11"},{"bbox":[106,515,176,528],"type":"text","angle":0,"index":12,"text":"如何计算 loss？","color":{"line":"rgba(13, 83, 222, 1)","fill":"rgba(13, 83, 222, 1)"},"page_idx":6,"id":"ac70c4e8-6fda-41be-96fc-d25ecb1b4388","page_size":[595,841],"block_position":"6-12"}]],"mergeConnections":[]}