A carregar...
Publicação
Improving malware detection with neuroevolution : a study with the semantic learning machine
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.84 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Machine learning has become more attractive over the years due to its remarkable adaptation and
problem-solving abilities. Algorithms compete amongst each other to claim the best possible results
for every problem, being one of the most valued characteristics their generalization ability.
A recently proposed methodology of Genetic Programming (GP), called Geometric Semantic Genetic
Programming (GSGP), has seen its popularity rise over the last few years, achieving great results
compared to other state-of-the-art algorithms, due to its remarkable feature of inducing a fitness
landscape with no local optima solutions. To any supervised learning problem, where a metric is used
as an error function, GSGP’s landscape will be unimodal, therefore allowing for genetic algorithms to
behave much more efficiently and effectively.
Inspired by GSGP’s features, Gonçalves developed a new mutation operator to be applied to the Neural
Networks (NN) domain, creating the Semantic Learning Machine (SLM). Despite GSGP’s good results
already proven, there are still research opportunities for improvement, that need to be performed to
empirically prove GSGP as a state-of-the-art framework.
In this case, the study focused on applying SLM to NNs with multiple hidden layers and compare its
outputs to a very popular algorithm, Multilayer Perceptron (MLP), on a considerably large classification
dataset about Android malware. Findings proved that SLM, sharing common parametrization with
MLP, in order to have a fair comparison, is able to outperform it, with statistical significance.
Descrição
Project Work presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
Palavras-chave
Geometric semantic genetic programming Artificial Neural Networks Genetic Programming Supervised Learning Semantic Learning Machine Multilayer Neural Networks