Verisig: verifying safety properties of hybrid systems with neural network controllers

Simple item page
dc.contributor.authorIvanov, Radoslav
dc.contributor.authorWeimer, James
dc.contributor.authorAlur, Rajeev
dc.contributor.authorPappas, George J.
dc.contributor.authorLee, Insup
dc.date2023-05-17T23:32:16.000
dc.date.accessioned2023-05-22T12:51:46Z
dc.date.available2023-05-22T12:51:46Z
dc.date.issued2019-04-01
dc.date.submitted2020-03-05T13:24:45-08:00
dc.description.abstractThis paper presents Verisig, a hybrid system approach to verifying safety properties of closed-loop systems using neural networks as controllers. We focus on sigmoid-based networks and exploit the fact that the sigmoid is the solution to a quadratic differential equation, which allows us to transform the neural network into an equivalent hybrid system. By composing the network’s hybrid system with the plant’s, we transform the problem into a hybrid system verification problem which can be solved using state-of-theart reachability tools. We show that reachability is decidable for networks with one hidden layer and decidable for general networks if Schanuel’s conjecture is true. We evaluate the applicability and scalability of Verisig in two case studies, one from reinforcement learning and one in which the neural network is used to approximate a model predictive controller.
dc.description.comments22nd ACM International Conference on Hybrid Systems: Computation and Control (HSCC 2019), Montreal, Canada, April, 16-18, 2019 (http://hscc2019.eecs.umich.edu/)
dc.identifier.urihttps://repository.upenn.edu/handle/20.500.14332/6940
dc.legacy.articleid1909
dc.legacy.fields10.1145/3302504.3311806
dc.legacy.fulltexturlhttps://repository.upenn.edu/cgi/viewcontent.cgi?article=1909&context=cis_papers&unstamped=1
dc.source.beginpage169
dc.source.endpage178
dc.source.issue859
dc.source.journalDepartmental Papers (CIS)
dc.source.journaltitle22nd ACM International Conference on Hybrid Systems: Computation and Control (HSCC 2019)
dc.source.peerreviewedtrue
dc.source.statuspublished
dc.subject.otherCPS Safe Autonomy
dc.subject.otherCPS Formal Methods
dc.subject.otherNeural Network Verification
dc.subject.otherHybrid Systems with Neural Network Controllers
dc.subject.otherLearning-Enabled Components
dc.subject.otherComputer Engineering
dc.subject.otherComputer Sciences
dc.titleVerisig: verifying safety properties of hybrid systems with neural network controllers
dc.typePresentation
digcom.identifiercis_papers/859
digcom.identifier.contextkey16726158
digcom.identifier.submissionpathcis_papers/859
digcom.typeconference
dspace.entity.typePublication
relation.isAuthorOfPublicationf7d2d91b-2a20-4596-9649-43b222476db4
relation.isAuthorOfPublication03b71a10-c612-4bb1-8a80-1463efee44c5
relation.isAuthorOfPublication0277bab3-1be2-4f99-9964-92e0ea36014e
relation.isAuthorOfPublication295f047c-e1ec-4b38-b5f4-0f30e7a34aea
relation.isAuthorOfPublication45a9eed5-3211-4c36-b40d-6394302dfdce
relation.isAuthorOfPublication.latestForDiscoveryf7d2d91b-2a20-4596-9649-43b222476db4
upenn.schoolDepartmentCenterDepartmental Papers (CIS)
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
verisig_scholarly.pdf
Size:
1.48 MB
Format:
Adobe Portable Document Format
Download
Collection
Presentations