Promises and challenges in applying deep learning to malware detection

Many different deep network architectures were proposed by machine learning practitioners and malware analysts to detect both known and unknown malware; some of the proposed architectures include restricted Boltzmann machines and hybrid methods. You can check some of them in the Further reading section. Novel approaches to detect malware and malicious software show many promising results. However, there are many challenges that malware analysts face when it comes to detecting malware using deep learning networks, especially when analyzing PE files because to analyze a PE file, we take each byte as an input unit, so we deal with classifying sequences with millions of steps, in addition to the need of keeping complicated spatial correlation across functions due to function calls and jump commands.

...