An Unsupervised Learning Model to Perform Side Channel Attack

摘要: This paper proposes a novel unsupervised learning approach for Power Analysis – form of side channel attack in Cryptanalysis. Different from existing works that exploit supervised framework to solve this problem, our method does not require any labeled pairs, which contains information the {X,Y}={key, power-trace}, but is still capable deciphering secret key accurately. Besides proposing regression-based, purpose, we further propose an enhanced model through exploiting dependency bits between different sub-processes during encryption process obtain accurate results more efficient way. Our experiment shows proposed outperforms state-of-the-art non-learning based decipherment methods significantly.