Dekompajliranje Java aplikacija potpomognuto metodama strojnog učenja

master's thesis

• Title (english): Decompilation of Java Applications Supported by Machine Learning Algorithms
• Author: Mihovil Kucijan
• Mentor: Stjepan Groš (mentor)
• Committee member: Stjepan Groš (predsjednik povjerenstva)
• Committee member: Ante Đerek (član povjerenstva)
• Committee member: Dejan Škvorc (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-07-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Razvojem strojnog učenja i povećanjem proceduralne moći računala, otvorio se novi način za pristupanju
problemima. Metode strojnog učenja kao neuronske mreže omogućavaju iskorištavanje velike količine
podataka kako bi se u sisteme ugradilo znanje o obrascima i pravilima bez direktnog programiranja u svrhu rješavanje problema.
U ovom radu sagledat će se moguća primjena umjetnih neuronskih mreža za probleme dekompajliranje Java bajtkoda. Prezentirana su dva modela koji nastoje otkriti strukturu
apstraktnih izraza izvornog jezika iz niza instrukcija koje su zapisane u izvršnom formatu.
Prvi model koristeći povratne neuronske mreže označava instrukcije koje su generirana iz
izraza uvjeta i petlja u izvornom kodu. Drugi model koristi povratne mreže kako bi iz
niza instrukcija generirao oznake početaka i kraja izraza izvornog jezika. Modeli
postižu minimalnu funkcionalnost prepoznavajući najjasnije obrasce izraza. Kod dužih
i kompliciranijih nizova otkrivaju tek dio strukture te nisu u mogućnosti naći jasne
granice izraza.

Abstract (english)

By development of machine learning techniques and increasing computer process abilities, new approach to solving problems has been presented. Machine learning methods like neural networks allow for analysis of large amounts of data to include knowledge of patterns and rules without usage of direct programming for solving problems.
This paper will consider application of artificial neural networks for the problems of decompiling Java bytecode. Two models have been introduced to find the structure of
abstract expressions of the original language from a series of instructions from a compiled
code.
The first model uses recurrent neural network to classify the instructions that were generated from
conditional and loop expressions in the source code. The other model uses a recurrent neural
networks to generate start and end labels of source expressions. Models
achieve minimal functionality by recognizing the clearest patterns of expressions. On longer
and more complicated array of instructions, model reveal only a part of the structure and is not able to find clear boundaries of expressions.

• Keywords: Java; Strojno učenje; Neuronske mreže; Dekompajliranje
• Keywords (english): Java; Machine Learning; Neural networks; Decompiling
• Language: croatian
• URN:NBN: urn:nbn:hr:168:633254
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-03-13 22:03:19