Using Variational Autoencoders to Increase the Performance of Malware Classification
Thomas Taylor, Amna Eleyan · 2021
Often complex datasets will have a large number of features for each of its samples. Sometimes, this can have a negative effect on the performance of models trained on the raw data. By reducing the number of features this problem can be avoided. However, this may cause a loss of information. One method to mitigate this is by using a type of unsupervised neural network structure called autoencoders. Autoencoders can be used to generate a reduced feature space which the models can then be trained on. This paper uses Convolutional Variational Autoencoders in order to create these latent features and then determines their effectiveness of improving performance of machine learning classifier models.