SCA_Uncertainty_Decomposition

This repository provides the implementation of the framework introduced in paper "Uncertainty Estimation in Neural Network-enabled Side-channel Analysis and Links to Explainability"

This paper introduces a unified framework for quantifying and explaining predictive uncertainty in neural network–based side-channel analysis. By leveraging matrix-based Rényi entropy and α-divergence, we decompose uncertainty into epistemic and aleatoric components and analyze how data quality, physical effects, and training choices impact key recovery. To localize the sources of uncertainty in side-channel traces, we integrate Shapley value–based explanations that identify time samples most responsible for unreliable predictions. Extensive experiments show that the proposed uncertainty measures strongly correlate with standard SCA metrics such as key rank, providing a complementary lens for evaluating attack effectiveness and complexity.

Dependencies

Install dependencies: tensorflow:

pip install tensorflow

scipy:

pip install scipy

h5py:

pip install h5py

Acknowledgements

1. This project uses code from the MRE repository by Shujian Yu for implementation of the MRE.
2. The codes from AutoSCA were used to train the models.
3. TCHES20V3_CNN_SCAPublic and Methodology-for-efficient-CNN-architectures-in-SCA were used for the model trained on ASCADf dataset.