Project description
Automated software that tests software systems like a person would
Even the simplest of computer programs can exhibit run-time errors when unexpected inputs are received. Testing software before release relies largely on humans trying a virtually endless list of actions that could lead to system failure. Automating just part of this process would significantly improve the efficiency and effectiveness of verification activities and slash development costs. Funded by the European Research Council, the AST project will develop a pre-commercial tool to efficiently and automatically generate semantically relevant system test cases. It will exploit reinforcement learning techniques to incrementally and automatically learn the most useful testing actions that can be performed on a graphical user interface.
Objective
Verifying the correctness of software systems requires extensive and expensive testing sessions. While there are tools and methodologies to efficiently address unit and integration testing, system testing is still largely the result of manual effort.
Testing software applications at the system level requires executing the applications through their interfaces to verify the correctness of their functionalities and stimulate all their layers and components. Automating just part of this process can dramatically improve the effectiveness of verification activities and significantly reduce development costs, relevantly alleviating developers from their verification effort.
This project addresses the development of a pre-commercial tool that has the unique capability of efficiently and automatically generating semantically-relevant system test cases equipped with functional oracles. This capability derives from the AUGUSTO technique, which is an outcome of the Learn ERC project. The idea behind Augusto is to exploit the common-sense knowledge, that is, the background knowledge that every computer user has and that normally lets her/him use software applications without the need of accessing any documentation or manual. Once this knowledge is represented abstractly and then embedded in AUGUSTO, the technique can automatically adapt its definition to the software under test every time a program is tested.
This development work will be performed jointly with A company that produces and markets testing tools.
Fields of science
- natural sciencescomputer and information sciencessoftwaresoftware applicationssystem software
- natural sciencescomputer and information sciencesknowledge engineering
- natural sciencescomputer and information sciencessoftwaresoftware development
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence
Programme(s)
Funding Scheme
ERC-POC - Proof of Concept GrantHost institution
20126 MILANO
Italy