I think this questions have two aspects to correctly answer. First, the detection rate is higher in the supervised learning than unspervised. Second, the unspervised can easily detect know and unkown attacks and this will be better than the supervised one. You, ultimatel, can apply ensemble deep learning to improve your scheme of detecting malware.

My answer is based on some questions. Why would you like to applied a deep learning concept for malware detection? Aren't there no better solutions? Have you check different solutions in respect to malware detection? What would be the advantages of DL concepts? What are the disadvantages?

@Volker Lohweg: Actually the state of the art in malware detection research is shifting towards the use of machine learning for malware detection. I do not know of any other ways better than the use of machine learning.

@Faisal Ali Garba: Machine learning (ML) includes a plethora of different learning techniques. ML is not deep architectures, or at least not only. When deciding the way to address a task, keep in mind the Occam's razor guidance. Perhaps, a deep architecture is not necessary. That's Volker Lohweg is suggesting you. Next, I also suggest you to well understand the difference between supervised and unsupervised in order to properly decide when apply a supervised learning strategy w.r.t. an unsupervised one.

Antonio Staiano: Okay but from the latest papers I have reviewed so far, they seem to be pointing towards the edge deep learning has over the traditional machine learning techniques. I have reviewed separate papers showing the similar detection result obtained using Deep Learning Multilayer Perceptron (supervised learning) and Deep Belief Network (unsupervised learning). That is why I asked with regards to malware detection scenario (of PE malwares) which would be best and why?