Side-channel analysis is one of the main threats against software and hardware implementations of cryptographic algorithms. Among those attacks, the profiling attacks are some of the most powerful though they present some weaknesses such as, for example, the need for synchronization between different execution of the algorithm. Recent years saw the use of deep neural networks to try to solve those weaknesses. Our work focused on trying to better understand the networks and how they make their predictions. This presentation will introduce some key concepts on which this research is based and then focus on understanding and exploring several results stemming from this work. Those results include a new method of performing the profiling attacks as well as the design of a new loss function derived from the Learning to Rank approach of the field of machine learning.