Direkt zum InhaltDirekt zur SucheDirekt zur Navigation
▼ Zielgruppen ▼

Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Probevortrag Promotion: Yannic Noller

Wann 02.04.2020 ab 11:00 (Europe/Berlin / UTC200) iCal
Wo online
Kontaktname

Am 2. April 2020 um 11.00 Uhr wird Herr Yannic Noller einen Vortrag zum Thema

“Hybrid Differential Software Testing”

halten. Mit diesem Vortrag möchte er Ihnen sein Promotionsthema vorstellen.

Aufgrund der aktuellen Situation werden wir den Probevortrag von Herrn Yannic Noller zum Thema "Hybrid Differential Software Testing” nun online durchführen.

Die notwendigen Instruktionen, wenn Sie sich zuschalten möchten:

Meeting Room Name:   Probevortrag Yannic Noller

Meeting Room Number: 97927303

Guest PIN: 60677

In order to join the meeting you can select one of the alternatives:

1. With your Browser under Windows, MacOS, GNU/Linux:

   https://conf.dfn.de/webapp/conference/97927303

2. With a SIP/H.323 room system or SIP/H.323 software client:

   H.323: 004910097927303

   SIP:   97927303@conf.dfn.de   

3. Telephone (audio only): Call 0049 30 200 97927303

4. Via Skype for Business: 97927303@vc.dfn.de

5. Using a mobile smart phone.

   You have to install the Pexip App beforehand:

   Android: https://play.google.com/store/apps/details?id=com.pexip.infinityconnect

   iOS: https://itunes.apple.com/us/app/pexip-infinity-connect/id1195088102

  Please Enter:   97927303@conf.dfn.de

Please wait for your host to join the meeting room.

For a connection test look at following website: https://www.conf.dfn.de/en/beschreibung-des-dienstes/zugangswege/test-call-service/

See our Privacy Policy to learn about your personal data and all your related rights and obligations: https://www.conf.dfn.de/datenschutz

 

Abstract

Differential software testing aims at the automated generation of test inputs that reveal behavioral differences in software. The search for behavioral differences can be separated into two categories in which the analysis executes: (1) different program versions with the same input, and (2) the same program with different inputs. Therefore, detecting regression bugs in software evolution, analyzing side-channels in programs, maximizing the execution cost of a program over multiple executions, and evaluating robustness in deep neural networks (DNNs) can all be seen as instances of differential software analysis, where the goal is to generate diverging executions of program paths.

The key challenge of differential software testing is to simultaneously reason about multiple program paths, often across program variants, in an efficient way.

The existing work on the various areas of differential testing show that the single techniques come with their own disadvantages which limits the exploration space of the analysis. An efficient and effective testing approach asks for a hybrid execution setup.

Therefore, this thesis proposes the concept of Hybrid Differential Software Testing (HyDiff): a hybrid solution combining search-based testing with a systematic exploration technique. HyDiff's search-based component leverages differential fuzzing directed by differential heuristics. HyDiff’s systematic exploration is based on dynamic symbolic execution which allows to incorporate concrete inputs in its analysis.

HyDiff is evaluated based on a quantitative analysis with experiments and benchmarks in the specific application scenarios of differential testing.