Software Engineering Patterns for Machine Learning Applications (SEP4MLA) - Part 5

Software Engineering Patterns for Machine Learning Applications (SEP4MLA) - Part 5 - Explainable Proxy Model. Washizaki, H., Khomh, F., Gu�h�neuc, Y., Takeuchi, H., Okuda, S., & Natori, N. In Vranić, V. & Brown, K., editors, Proceedings of the 30th Conference on Pattern Languages of Programs (PLoP), pages 1–10, October, 2023. ACM Press. 10 pages.

Paper abstract bibtex

Machine learning (ML) researchers and practitioners study the best practices to develop and support ML-based applications to ensure the quality and resolve constraints applied to their applications. Following best practices in software engineering, they often formalize these practices as software patterns. To clarify an overview of the current landscape of these practices and patterns, we discovered softwareengineering design patterns for machine-learning applications by thoroughly searching the available literature on the subject. Among the ML patterns found, we describe in this paper one ML topology pattern, “Explainable Proxy Model”, in the standard pattern format so that practitioners can (re)use it in their contexts and benefits from its advantages. The pattern addresses the problem of low explainability and reproducibility of highly accurate machine learning models by building a surrogate proxy ML model, called a canary model, which approximates the behavior of the best ML models (i.e., primary models) and monitoring deviations between canary and primary models. By describing the "Explainable Proxy Model" pattern, we make explicit its advantages and limitations as well as the contexts in which it applies.

@INPROCEEDINGS{Washizaki23-PLoP-SEP4MLAPart5,
   AUTHOR       = {Hironori Washizaki and Foutse Khomh and 
      Yann-Ga�l Gu�h�neuc and Hironori Takeuchi and Satoshi Okuda and 
      Naotake Natori},
   BOOKTITLE    = {Proceedings of the 30<sup>th</sup> Conference on Pattern Languages of Programs (PLoP)},
   TITLE        = {Software Engineering Patterns for Machine Learning 
      Applications (SEP4MLA) - Part 5 - Explainable Proxy Model},
   YEAR         = {2023},
   OPTADDRESS   = {},
   OPTCROSSREF  = {},
   EDITOR       = {Valentino Vranić and Kyle Brown},
   MONTH        = {October},
   NOTE         = {10 pages.},
   OPTNUMBER    = {},
   OPTORGANIZATION = {},
   PAGES        = {1--10},
   PUBLISHER    = {ACM Press},
   OPTSERIES    = {},
   OPTVOLUME    = {},
   KEYWORDS     = {Topic: <b>Design patterns</b>, Venue: <c>PLoP</c>},
   URL          = {http://www.ptidej.net/publications/documents/PLoP23.doc.pdf},
   ABSTRACT     = {Machine learning (ML) researchers and practitioners 
      study the best practices to develop and support ML-based applications 
      to ensure the quality and resolve constraints applied to their 
      applications. Following best practices in software engineering, they 
      often formalize these practices as software patterns. To clarify an 
      overview of the current landscape of these practices and patterns, we 
      discovered softwareengineering design patterns for machine-learning 
      applications by thoroughly searching the available literature on the 
      subject. Among the ML patterns found, we describe in this paper one 
      ML topology pattern, “Explainable Proxy Model”, in the standard 
      pattern format so that practitioners can (re)use it in their contexts 
      and benefits from its advantages. The pattern addresses the problem 
      of low explainability and reproducibility of highly accurate machine 
      learning models by building a surrogate proxy ML model, called a 
      canary model, which approximates the behavior of the best ML models 
      (i.e., primary models) and monitoring deviations between canary and 
      primary models. By describing the "Explainable Proxy Model" pattern, 
      we make explicit its advantages and limitations as well as the 
      contexts in which it applies.}
}

Downloads: 0

{"_id":"yXmAhhKRZWcprX6Da","bibbaseid":"washizaki-khomh-guhneuc-takeuchi-okuda-natori-softwareengineeringpatternsformachinelearningapplicationssep4mlapart5explainableproxymodel-2023","author_short":["Washizaki, H.","Khomh, F.","Gu�h�neuc, Y.","Takeuchi, H.","Okuda, S.","Natori, N."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"firstnames":["Hironori"],"propositions":[],"lastnames":["Washizaki"],"suffixes":[]},{"firstnames":["Foutse"],"propositions":[],"lastnames":["Khomh"],"suffixes":[]},{"firstnames":["Yann-Ga�l"],"propositions":[],"lastnames":["Gu�h�neuc"],"suffixes":[]},{"firstnames":["Hironori"],"propositions":[],"lastnames":["Takeuchi"],"suffixes":[]},{"firstnames":["Satoshi"],"propositions":[],"lastnames":["Okuda"],"suffixes":[]},{"firstnames":["Naotake"],"propositions":[],"lastnames":["Natori"],"suffixes":[]}],"booktitle":"Proceedings of the 30th Conference on Pattern Languages of Programs (PLoP)","title":"Software Engineering Patterns for Machine Learning Applications (SEP4MLA) - Part 5 - Explainable Proxy Model","year":"2023","optaddress":"","optcrossref":"","editor":[{"firstnames":["Valentino"],"propositions":[],"lastnames":["Vranić"],"suffixes":[]},{"firstnames":["Kyle"],"propositions":[],"lastnames":["Brown"],"suffixes":[]}],"month":"October","note":"10 pages.","optnumber":"","optorganization":"","pages":"1–10","publisher":"ACM Press","optseries":"","optvolume":"","keywords":"Topic: Design patterns, Venue: <c>PLoP</c>","url":"http://www.ptidej.net/publications/documents/PLoP23.doc.pdf","abstract":"Machine learning (ML) researchers and practitioners study the best practices to develop and support ML-based applications to ensure the quality and resolve constraints applied to their applications. Following best practices in software engineering, they often formalize these practices as software patterns. To clarify an overview of the current landscape of these practices and patterns, we discovered softwareengineering design patterns for machine-learning applications by thoroughly searching the available literature on the subject. Among the ML patterns found, we describe in this paper one ML topology pattern, “Explainable Proxy Model”, in the standard pattern format so that practitioners can (re)use it in their contexts and benefits from its advantages. The pattern addresses the problem of low explainability and reproducibility of highly accurate machine learning models by building a surrogate proxy ML model, called a canary model, which approximates the behavior of the best ML models (i.e., primary models) and monitoring deviations between canary and primary models. By describing the \"Explainable Proxy Model\" pattern, we make explicit its advantages and limitations as well as the contexts in which it applies.","bibtex":"@INPROCEEDINGS{Washizaki23-PLoP-SEP4MLAPart5,\r\n AUTHOR = {Hironori Washizaki and Foutse Khomh and \r\n Yann-Ga�l Gu�h�neuc and Hironori Takeuchi and Satoshi Okuda and \r\n Naotake Natori},\r\n BOOKTITLE = {Proceedings of the 30th Conference on Pattern Languages of Programs (PLoP)},\r\n TITLE = {Software Engineering Patterns for Machine Learning \r\n Applications (SEP4MLA) - Part 5 - Explainable Proxy Model},\r\n YEAR = {2023},\r\n OPTADDRESS = {},\r\n OPTCROSSREF = {},\r\n EDITOR = {Valentino Vranić and Kyle Brown},\r\n MONTH = {October},\r\n NOTE = {10 pages.},\r\n OPTNUMBER = {},\r\n OPTORGANIZATION = {},\r\n PAGES = {1--10},\r\n PUBLISHER = {ACM Press},\r\n OPTSERIES = {},\r\n OPTVOLUME = {},\r\n KEYWORDS = {Topic: Design patterns, Venue: <c>PLoP</c>},\r\n URL = {http://www.ptidej.net/publications/documents/PLoP23.doc.pdf},\r\n ABSTRACT = {Machine learning (ML) researchers and practitioners \r\n study the best practices to develop and support ML-based applications \r\n to ensure the quality and resolve constraints applied to their \r\n applications. Following best practices in software engineering, they \r\n often formalize these practices as software patterns. To clarify an \r\n overview of the current landscape of these practices and patterns, we \r\n discovered softwareengineering design patterns for machine-learning \r\n applications by thoroughly searching the available literature on the \r\n subject. Among the ML patterns found, we describe in this paper one \r\n ML topology pattern, “Explainable Proxy Model”, in the standard \r\n pattern format so that practitioners can (re)use it in their contexts \r\n and benefits from its advantages. The pattern addresses the problem \r\n of low explainability and reproducibility of highly accurate machine \r\n learning models by building a surrogate proxy ML model, called a \r\n canary model, which approximates the behavior of the best ML models \r\n (i.e., primary models) and monitoring deviations between canary and \r\n primary models. By describing the \"Explainable Proxy Model\" pattern, \r\n we make explicit its advantages and limitations as well as the \r\n contexts in which it applies.}\r\n}\r\n\r\n","author_short":["Washizaki, H.","Khomh, F.","Gu�h�neuc, Y.","Takeuchi, H.","Okuda, S.","Natori, N."],"editor_short":["Vranić, V.","Brown, K."],"key":"Washizaki23-PLoP-SEP4MLAPart5","id":"Washizaki23-PLoP-SEP4MLAPart5","bibbaseid":"washizaki-khomh-guhneuc-takeuchi-okuda-natori-softwareengineeringpatternsformachinelearningapplicationssep4mlapart5explainableproxymodel-2023","role":"author","urls":{"Paper":"http://www.ptidej.net/publications/documents/PLoP23.doc.pdf"},"keyword":["Topic: Design patterns","Venue: <c>PLoP</c>"],"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"http://www.yann-gael.gueheneuc.net/Work/Publications/Biblio/complete-bibliography.bib","dataSources":["8vn5MSGYWB4fAx9Z4"],"keywords":["topic: design patterns","venue: <c>plop</c>"],"search_terms":["software","engineering","patterns","machine","learning","applications","sep4mla","part","explainable","proxy","model","washizaki","khomh","gu�h�neuc","takeuchi","okuda","natori"],"title":"Software Engineering Patterns for Machine Learning Applications (SEP4MLA) - Part 5 - Explainable Proxy Model","year":2023}