Change search
ReferencesLink to record
Permanent link

Direct link
A Learning-based Approach to Unit Testing of Numerical Software
KTH, School of Computer Science and Communication (CSC), Theoretical Computer Science, TCS.ORCID iD: 0000-0002-9706-5008
KTH, School of Computer Science and Communication (CSC), Theoretical Computer Science, TCS.
2010 (English)In: 22nd IFIPInternational Conference on Testing Software and Systems, Natal, Brazil, Nov. 8-12, 2010 / [ed] Alexandre Petrenko, Adenilso da Silva Simão, José Carlos Maldonado, Berlin, Heidelberg: Springer , 2010, 221-235 p.Conference paper (Refereed)
Abstract [en]

We present an application of learning-based testing to theproblem of automated test case generation (ATCG) for numerical soft-ware. Our approach uses n-dimensional polynomial models as an algo-rithmically learned abstraction of the SUT which supports n-wise testing.Test cases are iteratively generated by applying a satisfiability algorithmto first-order program specifications over real closed fields and iterativelyrefined piecewise polynomial models.We benchmark the performance of our iterative ATCG algorithm againstiterative random testing, and empirically analyse its performance in find-ing injected errors in numerical codes. Our results show that for softwarewith small errors, or long mean time to failure, learning-based testing isincreasingly more efficient than iterative random testing.

Place, publisher, year, edition, pages
Berlin, Heidelberg: Springer , 2010. 221-235 p.
, Lecture Notes in Computer Science, ISSN 0302-9743 ; 6435
Keyword [en]
Machine learning, Software Testing, Model Checking
National Category
Computer Science
URN: urn:nbn:se:kth:diva-40870DOI: 10.1007/978-3-642-16573-3_16ISI: 000289226300016ScopusID: 2-s2.0-78649885303ISBN: 978-3-642-16572-6OAI: diva2:442572
Available from: 2011-09-26 Created: 2011-09-21 Last updated: 2011-10-12Bibliographically approved
In thesis
1. Learning-based Software Testing using Symbolic Constraint Solving Methods
Open this publication in new window or tab >>Learning-based Software Testing using Symbolic Constraint Solving Methods
2011 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

Software testing remains one of the most important but expensive approaches to ensure high-quality software today. In order to reduce the cost of testing, over the last several decades, various techniques such as formal verification and inductive learning have been used for test automation in previous research.

In this thesis, we present a specification-based black-box testing approach, learning-based testing (LBT), which is suitable for a wide range of systems, e.g. procedural and reactive systems. In the LBT architecture, given the requirement specification of a system under test (SUT), a large number of high-quality test cases can be iteratively generated, executed and evaluated by means of combining inductive learning with constraint solving.

We apply LBT to two types of systems, namely procedural and reactive systems. We specify a procedural system in Hoare logic and model it as a set of piecewise polynomials that can be locally and incrementally inferred. To automate test case generation (TCG), we use a quantifier elimination method, the Hoon-Collins cylindric algebraic decomposition (CAD), which is applied on only one local model (a bounded polynomial) at a time.

On the other hand, a reactive system is specified in temporal logic formulas, and modeled as an extended Mealy automaton over abstract data types (EMA) that can be incrementally learned as a complete term rewriting system (TRS) using the congruence generator extension (CGE) algorithm. We consider TCG for a reactive system as a bounded model checking problem, which can be further reformulated into a disunification problem and solved by narrowing.

The performance of the LBT frameworks is empirically evaluated against random testing for both procedural and reactive systems (executable models and programs). The results show that LBT is significantly more efficient than random testing in fault detection, i.e. less test cases and potentially less time are required than for random testing.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2011. vii, 47 p.
Trita-CSC-A, ISSN 1653-5723 ; 2011:15
Testing, Machine learning, Symbolic constraint solving, Model checking
National Category
Computer Science
urn:nbn:se:kth:diva-41932 (URN)978-91-7501-117-2 (ISBN)
2011-11-07, Sal F3, Lindstedtsvägen 26 KTH, Stockholm, 13:00 (English)
QC 20111012Available from: 2011-10-12 Created: 2011-10-03 Last updated: 2011-10-12Bibliographically approved

Open Access in DiVA

KMFN2010(855 kB)859 downloads
File information
File name FULLTEXT01.pdfFile size 855 kBChecksum SHA-512
Type fulltextMimetype application/pdf

Other links

Publisher's full textScopus final publication is available at

Search in DiVA

By author/editor
Meinke, KarlNiu, Fei
By organisation
Theoretical Computer Science, TCS
Computer Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 859 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Altmetric score

Total: 459 hits
ReferencesLink to record
Permanent link

Direct link