Hybrid genetic algorithm and deep learning techniques for advanced side-channel attacks.

Area of Science:

• Cryptography
• Computer Science
• Machine Learning

Background:

• Deep learning has advanced side-channel analysis (SCA).
• Hyperparameter optimization is crucial for effective neural network training in SCA.
• Conventional methods like grid search and Bayesian optimization have limitations in complex search spaces.

Purpose of the Study:

• To introduce a genetic algorithm (GA) framework for efficient hyperparameter optimization in deep learning-based SCA.
• To overcome the scalability issues of grid search and the high-dimensionality challenges of Bayesian optimization.
• To systematically identify optimal neural network configurations for enhanced SCA model performance.

Main Methods:

• Development of a genetic algorithm (GA) framework for hyperparameter search.
• Leveraging evolutionary strategies to navigate non-differentiable, multimodal optimization landscapes.
• Evaluation of the GA framework on protected Advanced Encryption Standard (AES) implementations.

Main Results:

• The GA-based approach achieved 100% key recovery accuracy on protected AES implementations.
• The GA significantly outperformed random search baselines (70% accuracy).
• Compared to Bayesian optimization, reinforcement learning, and tree-structured Parzen estimators, the GA achieved top performance in 25% of test cases and ranked second overall.

Conclusions:

• Genetic algorithms are a robust alternative for optimizing SCA models.
• The GA framework offers scalability and consistent performance across diverse attack scenarios.
• This research advances the assessment of cryptographic security through improved SCA techniques.