Reinforcement learning based hyper-parameter tuning in side-channel attacks

Keysight Device Security Lab, formerly Riscure, helps global leaders in semiconductor, mobile, media, automotive, and IoT secure their devices and embedded systems.

Deep learning models for SCA involve numerous hyperparameters (e.g., learning rates, layer sizes, batch sizes). Optimal configurations can significantly impact attack success but are usually searched via random or grid search, which is inefficient. Following [RWP+21], this topic applies reinforcement learning to automate hyperparameter tuning, aiming to reduce manual effort and improve attack reproducibility.

[RWP+21] Rijsdijk, J., Wu, L., Perin, G., & Picek, S. (2021). Reinforcement learning for hyperparameter tuning in deep learning-based side-channel analysis. IACR Transactions on Cryptographic Hardware and Embedded Systems, 677-707.

Responsibilities

• Learn about reinforcement learning strategies
• Customize reinforcement learning strategies in SCA model parameter and training parameter exploring
• Compare different strategies performance on benchmark datasets (symmetric and asymmetric SCA datasets)
• Compare reinforcement learning strategy with grid search strategy

Qualifications

• Pursuing a Bachelor’s degree, Master’s degree, or thesis project in Computer Science, Electrical Engineering, Cybersecurity, Embedded Systems, Computer Engineering, or a related field
• Have Knowledge about reinforcement learning and Keras framework
• Programming experience in Python
• Interest in embedded systems, hardware security, cybersecurity, cryptography, software security, IoT security, or security testing