DGA Detection Using Similarity-Preserving Bloom Encodings

Abstract

The sanitization of concise data samples can be challenging, as they do not provide a clear distinction between sensitive and non-sensitive parts within individual samples. In this context, traditional sanitization and anonymization measures are not applicable. We consider the detection of algorithmically generated domains through machine learning as an example of such a case, where the benign samples may leak sensitive information. Within this scenario, we evaluate the use of a similarity-preserving Bloom encoding technique to obscure the training samples.