Everything All at Once: Deep Learning Side-Channel Analysis Optimization Framework

Deep learning is becoming an increasingly proficient tool for side-channel analysis. While deep learning has been evolving around the tasks of image and speech recognition for decades, it is still lacking maturity for side-channel analysis. One of the challenges to train a good model is the fine-tuning of its hyperparameters. Many methods have been developed for Hyperparameter Optimization, but a few have been applied for deep learning side-channel analysis. We study the use of sampling algorithm and early-stopping mechanism in the hyperparameter optimization search for deep learning side-channel analysis models. We also offer a scalable deep learning framework to extend results obtained for other problems and datasets. Our results show that hyperparameter optimization methods can save time and resources while leading to models that can lead to the best possible output and at the same time are providing more confidence whether to look for a better model.