因果信息在不同粒度上的迁移性
|
摘要
知识与粒度相关,在不同粒度上对现象的解释不同,而因果性描述的是现象的本质特征。因果性与粒度之间存在着怎样的关联,一个粒度上的因果关系是否可移植到其他不同粒度上,是目前人工智能研究亟待解决的问题。针对由观测数据构成的信息系统,从数据中直接抽取因果变量所需满足的基本图形结构,估算变量间的因果关系;再通过向系统中添加新属性以及合并多个信息系统,改变原系统中信息的粒度,研究所识别的因果关系在新系统中的可迁移性。若新属性作用于结果变量,则原系统中的因果关系不可迁移至新系统;若新属性对结果变量无影响,则原系统中的因果关系可移植至新系统。
The knowledge we learned is grain-dependent,which leads to different explanations for a phenomena at different granularities.Causality characterizes the essence of the phenomena.These factors raise an urgent problem currently to be solved in artificial intelligence:the relationship between causality and granularity as well as the transportability of causal effect at one granularity over to a different granularity.Aiming at the information system gathered from observational data,the basic graphical structures required for causal variables can be extracted directly from the data.According to these structures,the causal effects between variables can be computed.By adding new attributes to system and merging multiple information systems,the granularity in the original system is changed and then the issue of whether the causal effect can be transported to the new system is settled in detail.The causal relationship from the original system cannot be transported to the new system if the new attribute acts on the effect variable,otherwise the transportability is feasible in the new system.
The knowledge we learned is grain-dependent,which leads to different explanations for a phenomena at different granularities.Causality characterizes the essence of the phenomena.These factors raise an urgent problem currently to be solved in artificial intelligence:the relationship between causality and granularity as well as the transportability of causal effect at one granularity over to a different granularity.Aiming at the information system gathered from observational data,the basic graphical structures required for causal variables can be extracted directly from the data.According to these structures,the causal effects between variables can be computed.By adding new attributes to system and merging multiple information systems,the granularity in the original system is changed and then the issue of whether the causal effect can be transported to the new system is settled in detail.The causal relationship from the original system cannot be transported to the new system if the new attribute acts on the effect variable,otherwise the transportability is feasible in the new system.
引文
[1] LAKE B M,ULLMAN T D,TENENBAUM J B,et al.Building machines that learnand think like people[J].Behavioral and Brain Sciences,2017,40:1-58.
[2] POGGIO T,SMALE S.The mathematics of learning:dealing with data[J].Notices of the American Mathematical Society,2003,50(5):537-544.
[3] HOBBS J R.Granularity[C]∥Proceedings of the 9th International Joint Conference on Artificial Intelligence.San Francisco:Morgan Kaufmann,1985:432-435.
[4]苗夺谦,李德毅,姚一豫,等.不确定性与粒计算[M].北京:科学出版社,2011.
[5] ZHANG X Y,MIAO D Q.Quantitative/qualitative regionchange uncertainty/certainty in attribute reduction:comparative region-change analysis based on granular computing[J].Information Sciences,2016,334-335:174-204.
[6] PAWLAK Z.Rough Sets:Theoretical Aspects of Reasoning about Data[M].Dordrecht:Kluwer Academic Publishers,1991.
[7] ZADEH L A.Toward a theory of fuzzy information granulation and its centrality in human reasoning and fuzzy logic[J].Fuzzy Sets and Systems,1997,90(90):111-127.
[8]张铃,张钹.问题求解理论及应用——商空间粒度计算理论及应用(第2版)[M].北京:清华大学出版社,2007.
[9] PEDRYCZ W.Granular Computing:Analysis and Design of Intelligent Systems[M].Florida:CRC Press,2013.
[10]PAWLAK Z.Granularity of knowledge,indiscernibility and rough sets[C]∥Proceedings of IEEE International Conference on Fuzzy Systems.IEEE,1998:106-110.
[11]YAO N,MIAO D Q,ZHANG Z F,et al.Probabilistic estimation forgeneralized rough modus ponens and rough modus tollens[C]∥Proceedings of International Joint Conference on Rough Sets.Springer International Publishing,2016:166-176.
[12]PEARL J.Causality:Models,Reasoning,and Inference(2nd ed)[M].New York:Cambridge University Press,2009.
[13]BENFERHAT S,SMAOUI S.Possibilistic causal networks for handling interventions:a new propagation algorithm[C]∥Proceedings of the 22nd AAAI Conference on Artificial Intelligence.AAAI Press,2007:373-378.
[14]BENFERHAT S.Interventions and belief change in possibilistic graphical models[J].Artificial Intelligence,2010,174(2):177-189.
[15]PAN S J,YANG Q.A survey on transfer learning[J].IEEE Transactions on Knowledge and Data Engineering,2010,22(10):1345-1359.
[16]ZUO H,ZHANG G,PEDRYCZ W,et al.Fuzzy regression transfer learning in Takagi-Sugenofuzzy models[J].IEEE Transactions on Fuzzy Systems,2017,25(6):1795-1807.
[17]PEARL J.What is gained from past learning[J].Journal of Causal Inference,2018,6(1):1-11.
[18]PEARL J,BAREINBOIM E.External validity:from do-calculus to transportability across populations[J].Statistical Science,2014,29(4):579-595.
[19]BAREINBOIM E,PEARL J.Transportability from multiple environments with limited experiments:completeness results[C]∥Advances in Neural Information Processing Systems 27.Curran Associates,2014:280-288.
[20]NGUYEN S H,BAZAN J,SKOWRON A,et al.Layered learning for concept synthesis[M]∥Transactions on Rough Sets I.Berlin:Springer,2004:187-208.
[21]BOUCHON-MEUNIER B,FOULLOY L,YAGER R R.Intelligent Systems for Information Processing-From Representation to Applications[M].Amsterdam:Elsevier Science B.V.,2003:243-252.
[22]WANG J,YAO Y Y,WANG F Y.“Rule+Exception”learning based on reduct[J].Chinese Journal of Computers,2005,28(11):1778-1789.(in Chinese)王珏,姚一豫,王飞跃.基于Reduct的“规则+例外”学习[J].计算机学报,2005,28(11):1778-1789.
[2] POGGIO T,SMALE S.The mathematics of learning:dealing with data[J].Notices of the American Mathematical Society,2003,50(5):537-544.
[3] HOBBS J R.Granularity[C]∥Proceedings of the 9th International Joint Conference on Artificial Intelligence.San Francisco:Morgan Kaufmann,1985:432-435.
[4]苗夺谦,李德毅,姚一豫,等.不确定性与粒计算[M].北京:科学出版社,2011.
[5] ZHANG X Y,MIAO D Q.Quantitative/qualitative regionchange uncertainty/certainty in attribute reduction:comparative region-change analysis based on granular computing[J].Information Sciences,2016,334-335:174-204.
[6] PAWLAK Z.Rough Sets:Theoretical Aspects of Reasoning about Data[M].Dordrecht:Kluwer Academic Publishers,1991.
[7] ZADEH L A.Toward a theory of fuzzy information granulation and its centrality in human reasoning and fuzzy logic[J].Fuzzy Sets and Systems,1997,90(90):111-127.
[8]张铃,张钹.问题求解理论及应用——商空间粒度计算理论及应用(第2版)[M].北京:清华大学出版社,2007.
[9] PEDRYCZ W.Granular Computing:Analysis and Design of Intelligent Systems[M].Florida:CRC Press,2013.
[10]PAWLAK Z.Granularity of knowledge,indiscernibility and rough sets[C]∥Proceedings of IEEE International Conference on Fuzzy Systems.IEEE,1998:106-110.
[11]YAO N,MIAO D Q,ZHANG Z F,et al.Probabilistic estimation forgeneralized rough modus ponens and rough modus tollens[C]∥Proceedings of International Joint Conference on Rough Sets.Springer International Publishing,2016:166-176.
[12]PEARL J.Causality:Models,Reasoning,and Inference(2nd ed)[M].New York:Cambridge University Press,2009.
[13]BENFERHAT S,SMAOUI S.Possibilistic causal networks for handling interventions:a new propagation algorithm[C]∥Proceedings of the 22nd AAAI Conference on Artificial Intelligence.AAAI Press,2007:373-378.
[14]BENFERHAT S.Interventions and belief change in possibilistic graphical models[J].Artificial Intelligence,2010,174(2):177-189.
[15]PAN S J,YANG Q.A survey on transfer learning[J].IEEE Transactions on Knowledge and Data Engineering,2010,22(10):1345-1359.
[16]ZUO H,ZHANG G,PEDRYCZ W,et al.Fuzzy regression transfer learning in Takagi-Sugenofuzzy models[J].IEEE Transactions on Fuzzy Systems,2017,25(6):1795-1807.
[17]PEARL J.What is gained from past learning[J].Journal of Causal Inference,2018,6(1):1-11.
[18]PEARL J,BAREINBOIM E.External validity:from do-calculus to transportability across populations[J].Statistical Science,2014,29(4):579-595.
[19]BAREINBOIM E,PEARL J.Transportability from multiple environments with limited experiments:completeness results[C]∥Advances in Neural Information Processing Systems 27.Curran Associates,2014:280-288.
[20]NGUYEN S H,BAZAN J,SKOWRON A,et al.Layered learning for concept synthesis[M]∥Transactions on Rough Sets I.Berlin:Springer,2004:187-208.
[21]BOUCHON-MEUNIER B,FOULLOY L,YAGER R R.Intelligent Systems for Information Processing-From Representation to Applications[M].Amsterdam:Elsevier Science B.V.,2003:243-252.
[22]WANG J,YAO Y Y,WANG F Y.“Rule+Exception”learning based on reduct[J].Chinese Journal of Computers,2005,28(11):1778-1789.(in Chinese)王珏,姚一豫,王飞跃.基于Reduct的“规则+例外”学习[J].计算机学报,2005,28(11):1778-1789.