From Passive to Active: Learning Timed Automata Efficiently

Bernhard Aichernig, Andrea Pferscher*, Martin Tappler

*Korrespondierende/r Autor/in für diese Arbeit

Publikation: KonferenzbeitragPaper

Abstract

Model-based testing is a promising technique for quality assurance. In practice, however, a model is not always present. Hence, model learning techniques attain increasing interest. Still, many learning approaches can only learn relatively simple types of models and advanced properties like time are ignored in many cases. In this paper we present an active model learning technique for timed automata. For this, we build upon an existing passive learning technique for real-timed systems. Our goal is to efficiently learn a timed system while simultaneously minimizing the set of training data. For evaluation we compared our active to the passive learning technique based on 43 timed systems with up to 20 locations and multiple clock variables. The results of 18060 experiments show that we require only 100 timed traces to adequately learn a timed system. The new approach is up to 755 times faster.
Originalspracheenglisch
Seiten1-19
Seitenumfang19
DOIs
PublikationsstatusVeröffentlicht - 10 Aug 2020
Veranstaltung12th NASA Formal Methods Symposium - NASA Ames Research Center, Moffett Field, USA / Vereinigte Staaten
Dauer: 12 Mai 202014 Mai 2020
https://ti.arc.nasa.gov/events/nfm-2020/

Konferenz

Konferenz12th NASA Formal Methods Symposium
KurztitelNFM 2020
LandUSA / Vereinigte Staaten
OrtMoffett Field
Zeitraum12/05/2014/05/20
Internetadresse

ASJC Scopus subject areas

  • !!Theoretical Computer Science
  • !!Computer Science(all)

Fields of Expertise

  • Information, Communication & Computing

Fingerprint Untersuchen Sie die Forschungsthemen von „From Passive to Active: Learning Timed Automata Efficiently“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren