×
蒂塔拉吉·达什;阿什温·斯里尼瓦桑;A.巴斯卡。

使用模式定向逆蕴涵将领域知识包含到GNN中。 （英语） Zbl 07510322号

机器。学习。 111，编号2，575-623（2022）。
摘要：我们提出了一种构建能够使用多关系领域知识的图形神经网络（GNN）的通用技术。该技术基于归纳逻辑编程（ILP）中开发的模式定向逆蕴涵（MDIE）。给定一个数据实例\（e）和背景知识\（B），MDIE确定一个最具体的逻辑公式\（bot_B（e）\），其中包含\（B \）中与\（e \）相关的所有关系信息。我们用“bottom-graph”表示\（\bot_B（e）\），它可以转换为适合GNN实现的形式。这种转换提供了一种将通用背景知识合并到GNN中的原则方法：我们使用术语“BotGNN”来表示这种形式的图形神经网络。对于几个GNN变体，使用具有大量背景知识的真实世界数据集，我们表明BotGNN的性能显著优于两个没有背景知识的GNN以及最近提出的将领域知识包含到GNN中的简化技术。我们还提供了实验证据，将BotGNN和使用表示“命题化”形式背景知识的特征的多层感知器进行了有利的比较；和BotGNN到基于使用最具体子句的标准ILP。综上所述，这些结果表明，BotGNN能够将GNN的计算效率与ILP的代表性多功能性结合起来。

MSC公司：

68T05型 人工智能中的学习和自适应系统

关键词：

神经符号学习;归纳逻辑程序设计;模式定向逆蕴涵;图形神经网络;背景知识

软件：

喀拉拉邦;千对数;克利夫兰;PyTorch公司;GNMT公司;CLEVR数据集;ImageNet公司;亚当;github;阿勒夫;数据链路2;AlexNet公司;WaveNet公司
PDF格式BibTeX公司 XML格式引用
\textit{T.Dash}等人，马赫。学习。111，编号2，575--623（2022；兹bl 07510322）
全文： 内政部 arXiv公司

参考文献：

[1] 安藤，HY；Dehaspe，L。；卢伊滕，W。；Van Craenenbroeck，E。；Vandecasteel，H。；Van Meervelt，L.，用归纳逻辑编程发现晶体中的氢键规则，分子药剂学，3，6，665-674（2006）·doi:10.1021/mp060034z
[2] Bai，S。；张，F。；Torr，PH，超图卷积和超图注意，模式识别，110，107637（2021）·doi:10.1016/j.patcog.2020.107637
[3] Besold，T.R.、Garcez，A.D.、Bader，S.、Bowman，H.、Domingos，P.、Hitzler，P.，Kühnberger，K.U.、Lamb，L.C.、Lowd，D.、Lima，P.M.V.等人（2017年）。神经符号学习和推理：一项调查和解释。arXiv:abs/1711.03902
[4] Bianchi，F.M.、Grattarola，D.、Livi，L.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Alippi，C.（2021）。带卷积arma滤波器的图形神经网络。IEEE模式分析和机器智能汇刊，第1-1页。doi:10.1109/TPAMI.2021.3054830
[5] Bravo，H.C.、Page，D.、Ramakrishnan，R.、Shavlik，J.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Costa，V.S.（2005）。归纳逻辑程序设计中面向集的计算框架及其在推广逆蕴涵中的应用。在归纳逻辑编程国际会议上，Springer，第69-86页·Zbl 1134.68467号
[6] Cangea，C.、Velićković，P.、Jovanović，N.、Kipf，T.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Lió，P.（2018）。面向稀疏层次图分类器。arXiv:abs/1811.01287
[7] Cholet，F.等人（2015年）。凯拉斯。https://keras.io。
[8] Dash，T.、Srinivasan，A.、Vig，L.、Orhobor，O.I.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；King，R.D.（2018）。深层关系机器的大规模评估。在归纳逻辑编程国际会议上，Springer，第22-37页。
[9] Dash，T.、Srinivasan，A.、Joshi，R.S.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Baskar，A.（2019年）。离散随机搜索及其在深关系机器特征选择中的应用。在国际人工神经网络会议上，施普林格，第29-45页。
[10] Dash，T.、Chitlangia，S.、Ahuja，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Srinivasan，A.（2021a）。如何告诉深层神经网络我们所知道的。arXiv:abs/2107.10295
[11] Dash，T.、Chitlangia，S.、Ahuja，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Srinivasan，A.（2021b）。将领域知识纳入深度神经网络。arXiv:abs/2103.00180
[12] Dash，T.、Srinivasan，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；维格·L（2021c）。将符号域知识纳入图形神经网络。机器学习，1-28·Zbl 07465651号
[13] 杜，SS；Hou，K。；Salakhutdinov，RR；波佐斯，B。；王，R。；Xu，K.，图神经切线核：图神经网络与图核的融合，神经信息处理系统进展，32，5723-5733（2019）
[14] Feng，Y。；你，H。；张，Z。；吉·R。；Gao，Y.，超图神经网络，AAAI人工智能会议论文集，333558-3565（2019）·doi:10.1609/aaai.v33i01.33013558
[15] Fey，M.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Lenssen，J.E.（2019年）。使用PyTorch Geometric进行快速图形表示学习。在ICLR关于图和流形表示学习的研讨会上。
[16] Fischer，M.、Balunovic，M.，Drachsler-Cohen，D.、Gehr，T.、Zhang，C.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Vechev，M.（2019）。用逻辑训练和查询神经网络。在机器学习国际会议上，PMLR，第1931-1941页。
[17] 弗朗索阿，MV；Zaverucha，G。；Garcez，ASd，使用人工神经网络的底子句命题化进行快速关系学习，机器学习，94，1，81-104（2014）·doi:10.1007/s10994-013-5392-1
[18] Frasconi，P。；Costa，F。；De Raedt，L。；De Grave，K.，klog：一种使用内核进行逻辑和关系学习的语言，《人工智能》，217，117-143（2014）·Zbl 1405.68288号 ·doi:10.1016/j.artint.2014.08.003
[19] Garcez，AD公司；戈里，M。；兰姆，LC；塞拉菲尼，L。；Spranger，M。；Tran，SN，《神经符号计算：机器学习和推理原理集成的有效方法》，FLAP，6,4611-632（2019）
[20] ASA加塞兹；Zaverucha，G.，《联结主义归纳学习和逻辑编程系统》，应用智能，11，1，59-77（1999）·doi:10.1023/A:1008328630915
[21] Gilmer，J.、Schoenholz，S.S.、Riley，P.F.、Vinyals，O.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Dahl，G.E.（2017）。量子化学的神经信息传递。在机器学习国际会议上，PMLR，第1263-1272页。
[22] Gori，M.、Monfardini，G.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Scarselli，F.（2005）。一种新的图域学习模型。诉讼中。2005年IEEE神经网络国际联合会议，2005年，IEEE，第2卷，第729-734页。
[23] Hamilton，W.、Ying，Z.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Leskovec，J.（2017）。大型图的归纳表示学习。《神经信息处理系统进展》，第1024-1034页。
[24] Hamilton，WL，图表示学习，人工智能和机器学习综合讲座，14，3，1-159（2020）·Zbl 1457.68005号 ·doi:10.2200/S01045ED1V01Y202009AIM046
[25] Heckerman，D.、Meek，C.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Koller，D.（2007）。概率实体关系模型、prms和板块模型。统计关系学习导论，第201-238页。
[26] Jankovics，V.（2020年）。瓦克尔/西尔普。https://github.com/vakker/CILP网站
[27] Johnson，J.、Hariharan，B.、van der Maaten，L.、Fei Fei，L.、Lawrence Zitnick，C.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Girshick，R.（2017）。Clevr：用于合成语言和基本视觉推理的诊断数据集。《IEEE计算机视觉和模式识别会议论文集》，第2901-2910页。
[28] Kersting，K.、Kriege，N.M.、Morris，C.、Mutzel，P.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Neumann，M.（2016）。图形内核的基准数据集。http://graphkernels.cs.tu-dortmund.de
[29] 金，RD；罗兰，J。；Oliver，秘书长；Young先生。；奥布里，W。；伯恩，E。；Liakata，M。；马克汉姆，M。；Pir，P。；索尔达托娃，LN，科学自动化，科学，324592385-89（2009）·doi:10.1126/science.1165620
[30] Kingma，D.P.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Ba，J.（2015）。亚当：一种随机优化方法。在ICLR（海报）中，arXiv:1412.6980
[31] Kipf，T.N.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Welling，M.（2017）。图卷积网络半监督分类。在2017年4月24日至26日于法国土伦举行的第五届国际学习代表大会上，《2017年ICLR会议记录》。
[32] Kitano，H.，《赢得诺贝尔奖及以后的人工智能：创造科学发现的引擎》，AI杂志，37，1，39-49（2016）·doi:10.1609/aimag.v37i1.2642
[33] Krizhevsky，A。；Sutskever，I。；Hinton，GE，Imagenet分类与深度卷积神经网络，ACM通信，60，6，84-90（2017）·doi:10.1145/3065386
[34] Kursuncu，U.、Gaur，M.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Sheth，A.（2020年）。知识注入学习（k-il）：将知识深度融入深度学习。arXiv:abs/1912.00512
[35] Lamb，L.C.、Garcez，A.D.、Gori，M.、Prates，M.O.、Avelar，P.H.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Vardi，M.Y.（2020年）。图形神经网络与神经符号计算：综述与展望。C.Bessiere（Ed.）《第二十九届国际人工智能联合会议论文集》，IJCAI-20，国际人工智能组织联合会议，第4877-4884页。doi:10.24963/ijcai.2020/679，测量轨道。
[36] Lavrać，N.、Díeroski，S.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Grobelnik，M.（1991年）。学习与莱纳斯关系的非递归定义。在欧洲学习工作会议上，施普林格，第265-281页。
[37] 拉夫拉奇，N。；波德佩坎，V。；Robnik-Šikonja，M.，关系数据的命题化，83-105（2021），Cham:Springer，Cham·Zbl 1466.68004号 ·数字对象标识代码：10.1007/978-3-030-68817-2-4
[38] Lee，J.、Lee，I.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Kang，J.（2019）。自我关注图形池。在机器学习国际会议上，第3734-3743页。
[39] Lodhi，H.（2013）。深层关系机器。在柏林施普林格神经信息处理国际会议上，第212-219页。
[40] 马克思·K·A、奥尼尔·P、霍夫曼·P；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Ujwal，M.（2003）。数据挖掘nci癌细胞系化合物gi50值：确定醌亚型对黑色素瘤和白血病细胞类别有效。化学信息与计算机科学杂志，43（5），1652-1667。
[41] 莫里斯，C。；Ritzert，M。；费伊，M。；汉密尔顿，WL；Lenssen，JE；藤条，G。；Grohe，M.，Weisfeiler和leman go neural:高阶图神经网络，《AAAI人工智能会议论文集》，33，4602-4609（2019）·doi:10.1609/aaai.v33i01.33014602
[42] Muggleton，S.，《逆蕴涵与程序》，新一代计算，13，3-4，245-286（1995）·doi:10.1007/BF03037227
[43] 麻格尔顿，S。；De Raedt，L.，《归纳逻辑编程：理论和方法》，《逻辑编程杂志》，19629-679（1994）·Zbl 0816.68043号 ·doi:10.1016/0743-1066（94）90035-3
[44] Muralidhar，N.、Islam，M.R.、Marwah，M.、Karpatne，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Ramakrishnan，N.（2018年）。将先验领域知识纳入深度神经网络。2018年IEEE国际大数据会议，IEEE，第36-45页。
[45] Oord，A.V.D.、Dieleman，S.、Zen，H.、Simonyan，K.、Vinyals，O.、Graves，A.、Kalchbrenner，N.、Senior，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Kavukcuoglu，K.（2016）。Wavenet：原始音频的生成模型。arXiv:abs/1609.03499
[46] Paszke，A.、Gross，S.、Massa，F.、Lerer，A.、Bradbury，J.、Chanan，G.、Killeen，T.、Lin，Z.、Gimelshein，N.、Antiga，L.等人（2019年）。Pytorch：一个命令式、高性能的深度学习库。《神经信息处理系统进展》，第8024-8035页。
[47] Plotkin，G.（1972年）。归纳推理的自动方法。爱丁堡大学博士论文。
[48] 普洛金，GD，关于归纳概括的注释，机器智能，5，1，153-163（1970）·兹伯利0219.68045
[49] Prechelt，L.（1998年）。提前停止——但什么时候？《神经网络：贸易的诡计》，柏林斯普林格出版社，第55-69页。
[50] Saha，A.、Srinivasan，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Ramakrishnan，G.（2012年）。什么样的关系特征对统计学习有用？在归纳逻辑编程国际会议上，Springer，第209-224页·Zbl 1382.68203号
[51] Schlichtkrull，M.、Kipf，T.N.、Bloem，P.、Van Den Berg，R.、Titov，I.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Welling，M.（2018）。用图卷积网络建模关系数据。在欧洲语义网络会议上，Springer，第593-607页。
[52] Sheth，A。；高尔，M。；Kursuncu，美国。；Wickramarachchi，R.，《加强深度学习的知识融合学习阴影》，IEEE互联网计算，23，6，54-63（2019）·doi:10.1109/MIC.2019.2960071
[53] Sourek，G。；Aschenbrenner，V。；Zelezny，F。；肖克尔特，S。；Kuzelka，O.，《提升的关系神经网络：潜在关系结构的有效学习》，《人工智能研究杂志》，62，69-100（2018）·Zbl 1444.68163号 ·doi:10.1613/jair.1.11203
[54] Šourek，G.，eleznỳ, F.、&amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Kuželka，O.（2021）。超越图神经网络和提升关系神经网络。机器学习，第1-44页·Zbl 1515.68279号
[55] Srinivasan，A.（2001）。aleph手册。https://www.cs.ox.ac.uk/activities/程序导入/Aleph/Aleph.html
[56] Srinivasan，A.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Ramakrishnan，G.（2011年）。使用设计的实验对ilp进行参数筛选和优化。机器学习研究杂志，12（2）·兹比尔1280.68197
[57] Srinivasan，A.、King，R.D.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Bain，M.E.（2003）。归纳逻辑程序设计中相关信息使用的实证研究。机器学习研究杂志，4（7月）：369-383·Zbl 1102.68597号
[58] 北斯利瓦斯塔瓦。；辛顿，G。；Krizhevsky，A。；Sutskever，I。；Salakhutdinov，R.，《辍学：防止神经网络过度拟合的简单方法》，《机器学习研究杂志》，1929-1958年，第15期，第1期（2014年）·Zbl 1318.68153号
[59] Stevens，R.、Taylor，V.、Nichols，J.、Maccabe，A.B.、Yelick，K.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Brown，D.（2020年）。爱科学。技术代表，美国伊利诺伊州阿贡市阿贡国家实验室（ANL）。
[60] Van Craenenbroeck，E.、Vandecasteele，H.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Dehaspe，L.（2002）。Dmax的功能组和环库。https://dtai.cs.kuleuven.be/software/dmax网站/
[61] Velic̆ković，P.，Cucurull，G.，Casanova，A.，Romero，A.，Lió，P.&amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Bengio，Y.（2018年）。绘制注意力网络图。在学习代表国际会议上，https://openreview.net/forum？id=rJXMpikCZ
[62] SVN维什瓦纳坦；Schraudolph，NN；康多，R。；博格沃德，KM，图核，机器学习研究杂志，11201-1242（2010）·Zbl 1242.05112号
[63] Wang，H.，Zhao，M.，Xie，X.，Li，W.，&amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；郭明（2019）。推荐系统的知识图卷积网络。在《全球网络会议》中，第3307-3313页。
[64] 威廉姆斯，K。；比尔斯兰，E。；斯帕克斯，A。；奥布里，W。；杨，M。；索尔达托娃，LN；De Grave，K。；拉蒙，J。；德·克莱尔，M。；Sirawaraporn，W.，《药物重新定位对抗被忽视的热带病验证了药物开发的更快更便宜》，《皇家学会界面杂志》，12，104，20141289（2015）·doi:10.1098/rsif.2014.1289
[65] Wu，Y.，Schuster，M.，Chen，Z.，Le，Q.V.，Norouzi，M.、Macherey，W.、Krikun，M.和Cao，Y.、Gao，Q.、Macheree，K.等（2016）。谷歌的神经机器翻译系统：弥合人类和机器翻译之间的鸿沟。arXiv:abs/1609.0814
[66] Wu，Z.，Pan，S.，Chen，F.，Long，G.，Zhang，C.，&amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Philip，S.Y.（2020）。图形神经网络综述。IEEE神经网络和学习系统汇刊。
[67] Xie，Y.，Xu，Z.，Kankanhalli，M.S.，Meel，K.S.，&amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Soh，H.（2019）。将符号知识嵌入深层网络。《神经信息处理系统进展》，第4233-4243页。
[68] Xu，J.，Zhang，Z.，Friedman，T.，Liang，Y.，&；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Broeck，G.（2018）。用于符号知识深度学习的语义损失函数。在机器学习国际会议上，PMLR，第5502-5511页。
[69] Xu，K.，Hu，W.，Leskovec，J.和；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；amp；Jegelka，S.（2019年）。图形神经网络有多强大？在学习代表国际会议上，https://openreview.net/forum？id=ryGs6iA5Km
此参考列表基于出版商或数字数学图书馆提供的信息。其项与zbMATH标识符进行启发式匹配，可能包含数据转换错误。在某些情况下，zbMATH Open的数据对这些数据进行了补充/增强。这试图尽可能准确地反映原始论文中列出的参考文献，而不要求完整或完全匹配。