Focus is Key to Success: A Focal Loss Function for Deep Learning-Based Side-Channel Analysis

Abstract

The deep learning-based side-channel analysis represents one of the most powerful side-channel attack approaches. Thanks to its capability in dealing with raw features and countermeasures, it becomes the de facto standard approach for the SCA community. The recent works significantly improved the deep learning-based attacks from various perspectives, like hyperparameter tuning, design guidelines, or custom neural network architecture elements. Still, insufficient attention has been given to the core of the learning process -the loss function.This paper analyzes the limitations of the existing loss functions and then proposes a novel side-channel analysis-optimized loss function: Focal Loss Ratio (FLR), to cope with the identified drawbacks observed in other loss functions. To validate our design, we 1) conduct a thorough experimental study considering various scenarios (datasets, leakage models, neural network architectures) and 2) compare with other loss functions used in the deep learning-based side-channel analysis (both "traditional" ones and those designed for side-channel analysis). Our results show that FLR loss outperforms other loss functions in various conditions while not having computational overhead like some recent loss function proposals.

Deep Neural Networks for external corrosion classification in industrial above-ground storage tanks

“…Nevertheless, it is not necessarily associated with more capabilities of generalization of the model. When using an alpha value of 0.25 the training focused the most on the minority class of the dataset, in our case, the blistering class; but paying less attention to the most abundant class, pitting [27]. However, if the performance of the model is stuck due to the majority class, this loss function, with these hyperparameters, does not offer the best solution for the desired task.…”

Tipping the Balance: Imbalanced Classes in Deep-Learning Side-Channel Analysis

“…The proposed loss function is a good alternative to standard loss functions in deep-learningbased SCA for imbalanced data. Kerkhof et al [10] continued the direction of the design of custom loss functions for SCA. The proposed loss function, the focal loss ratio, was designed to enable deep-learning models to learn from noisy or imbalanced data efficiently.…”

NeuroSCA: Evolving Activation Functions for Side-Channel Analysis

“…Zheng et al proposed a new metric function called Cross Entropy Ratio (CER), which they adapted to a new loss function specifically designed for deep learning in SCA [37]. Kerkhof et al proposed a loss function that is designed for SCA, performs well, and has low computational overhead [38]. Finally, the same authors also conducted a systematic analysis of custom loss functions and commonly used ones and concluded that custom loss functions indeed perform better in SCA [39].…”

“…• In [38], Kerkhof et al design custom loss functions for SCA. The training set size equals 50 000 traces, and the authors report median performance.…”