The SZTE Innovation Award has been awarded for five years to support researchers and students of the university who have developed a scientific result, an innovative solution that is intellectual and business-oriented. can be utilized. The most recently won Innovation Awards were presented in 5 categories. Viktor Csuvik received the “Most Innovative TDK Work” award for his work entitled “Further Development of Automatic Bug Repair Options”. Supervisor Dr. László Vidács
When we talk about an error in the framework of automatic program correction, it means some unexpected behavior of the program. Most of the time, these are not errors like “I press the green light button and the light is red when it should be green,” but rather some more complex relationship tends to trigger the faulty behavior. What happens to software in which the error is found? They usually outsource the task to software developers to fix it. There is nothing wrong with that, but why not automate the process? – Viktor Csuvik asked himself.
“During automatic program repair, the first step is to locate the source of the error. For the sake of simplicity, let’s say where we know the source of the error, and then the genetic algorithm that tries to correct the error comes into play.The genetic algorithm is motivated by biology, the methods used in it are related to it. , in which each individual is a variant of the original program (a descendant that differs from it in a small modification) As we progress over time, crossbreeding and mutation operators in the population produce newer individuals whose “goodness” is determined by the so-called fitness function. I think it can be felt, without further technical information, that the improvement depends largely on the success of using genetic operators on individuals , which really correct the original error, and to define the goodness of an individual as accurately as possible “- Viktor Csuvik summed up the essence of the research.
The essence of his approach is to strive for the simplicity of the fixes made, so that it is as close as possible to a program created by a developer. The following publication is also named after Viktor Csuvik, which reveals further interesting details on the topic, such as how to define whether a program is simple or not.
The tool was tested with 127 real program errors, of which A fix was made for 24. Of these 24 fixes, 5 are completely identical to a developer fix. This means that it is indistinguishable whether the developer wrote the patch or it was made automatically, 7 and its “behavior” is exactly the same. “On average, a fix was made in 1.5 hours, which may seem like a lot at first glance, but considering that we would otherwise have to pay a software developer’s salary, so only the electricity cost the algorithm ran is quite different. Our important result is also. in order to validate the corrections made, we did so in 50% of the cases, which was manifested in the fact that the corrections deemed good by the original genetic algorithm were automatically checked and any errors were filtered out, “the winner summed up.
Hardware, software, tests, curiosities and colorful news from the world of IT by clicking here
