Side-Channel Data Preprocessing using Enhanced Convolutional Autoencoders
In recent years,deep learning-based side-channel analysis methods have demonstrated significant potential and advantages.However,most of the current deep learning techniques are primarily focused on solving side-channel in-formation classification problems.To further enhance the effectiveness of side-channel attacks,this paper employs an im-proved convolutional autoencoder for preprocessing the ASCAD dataset.Through the dimensionality reduction process of the convolutional autoencoder,the ASCAD data exhibits a notable improvement in normalized inter-class variance(NICV)leakage.To verify the preprocessing effectiveness,we selected an MLP attack model to assess the attack per-formance on the preprocessed dataset.The experiments confirmed that our convolutional autoencoder-based preprocessing outperforms the traditional PCA preprocessing,yielding superior performance in side-channel attacks.These results un-derscore the effectiveness and superiority of the convolutional autoencoder in side-channel data preprocessing.
万方数据
维普
