Näytä suppeat kuvailutiedot

dc.contributor.authorMyllyaho, Lalli
dc.contributor.authorNurminen, Jukka K.
dc.contributor.authorMikkonen, Tommi
dc.date.accessioned2022-07-01T11:25:26Z
dc.date.available2022-07-01T11:25:26Z
dc.date.issued2022
dc.identifier.citationMyllyaho, L., Nurminen, J. K., & Mikkonen, T. (2022). Node co-activations as a means of error detection : Towards fault-tolerant neural networks. <i>Array</i>, <i>15</i>, Article 100201. <a href="https://doi.org/10.1016/j.array.2022.100201" target="_blank">https://doi.org/10.1016/j.array.2022.100201</a>
dc.identifier.otherCONVID_147286934
dc.identifier.urihttps://jyx.jyu.fi/handle/123456789/82179
dc.description.abstractContext: Machine learning has proved an efficient tool, but the systems need tools to mitigate risks during runtime. One approach is fault tolerance: detecting and handling errors before they cause harm. Objective: This paper investigates whether rare co-activations – pairs of usually segregated nodes activating together – are indicative of problems in neural networks (NN). These could be used to detect concept drift and flagging untrustworthy predictions. Method: We trained four NNs. For each, we studied how often each pair of nodes activates together. In a separate test set, we counted how many rare co-activations occurred with each input, and grouped the inputs based on whether its classification was correct, incorrect, or whether its class was absent during training. Results: Rare co-activations are much more common in inputs from a class that was absent during training. Incorrectly classified inputs averaged a larger number of rare co-activations than correctly classified inputs, but the difference was smaller. Conclusions: As rare co-activations are more common in unprecedented inputs, they show potential for detecting concept drift. There is also some potential in detecting single inputs from untrained classes. The small difference between correctly and incorrectly predicted inputs is less promising and needs further research.en
dc.format.mimetypeapplication/pdf
dc.language.isoeng
dc.publisherElsevier
dc.relation.ispartofseriesArray
dc.rightsCC BY 4.0
dc.subject.othermachine learning
dc.subject.otherfault tolerance
dc.subject.otherneural networks
dc.subject.othererror detection
dc.subject.otherconcept drift
dc.subject.otherdependability
dc.titleNode co-activations as a means of error detection : Towards fault-tolerant neural networks
dc.typearticle
dc.identifier.urnURN:NBN:fi:jyu-202207013778
dc.contributor.laitosInformaatioteknologian tiedekuntafi
dc.contributor.laitosFaculty of Information Technologyen
dc.type.urihttp://purl.org/eprint/type/JournalArticle
dc.type.coarhttp://purl.org/coar/resource_type/c_2df8fbb1
dc.description.reviewstatuspeerReviewed
dc.relation.issn2590-0056
dc.relation.volume15
dc.type.versionpublishedVersion
dc.rights.copyright© 2022 the Authors
dc.rights.accesslevelopenAccessfi
dc.subject.ysoneuroverkot
dc.subject.ysokoneoppiminen
dc.subject.ysovirheet
dc.subject.ysoluotettavuus
dc.format.contentfulltext
jyx.subject.urihttp://www.yso.fi/onto/yso/p7292
jyx.subject.urihttp://www.yso.fi/onto/yso/p21846
jyx.subject.urihttp://www.yso.fi/onto/yso/p148
jyx.subject.urihttp://www.yso.fi/onto/yso/p1629
dc.rights.urlhttps://creativecommons.org/licenses/by/4.0/
dc.relation.doi10.1016/j.array.2022.100201
jyx.fundinginformationhis work was funded by local authorities (“Business Finland”) under grant agreement ITEA-2019-18022-IVVES of ITEA3 programme and grant agreement ITEA-2020-20219-IML4E of ITEA4 programme .
dc.type.okmA1


Aineistoon kuuluvat tiedostot

Thumbnail

Aineisto kuuluu seuraaviin kokoelmiin

Näytä suppeat kuvailutiedot

CC BY 4.0
Ellei muuten mainita, aineiston lisenssi on CC BY 4.0