Evaluation and training method of deep learning model for side channel attack

Abstract

The invention discloses an evaluation and training method of a deep learning model for side channel attacks. Evaluation indexes can well reflect the condition of the model side channel attack effect, and the calculation complexity and the requirement for the ability of an evaluator are far lower than those of traditional side channel evaluation indexes. Besides, the evaluation indexes are transformed into a loss function for deep learning side channel attack, so that the problem of data imbalance which is widely faced by a deep learning side channel can be thoroughly solved under the condition that a data set and a network structure are not changed at all. When unbalanced data is processed, the effect of the loss function under various conditions is greatly superior to that of the conventional loss function.

Description

technical field [0001] The invention relates to the technical field of side channel attacks, in particular to an evaluation and training method of a deep learning model for side channel attacks. Background technique [0002] Since the concept of Side Channel Attack (SCA) was proposed in 1996, it has received extensive attention in the field of cryptography. Since the side channel information generated when the cryptographic algorithm is running on the cryptographic device can be used, including the energy consumption of the device, electromagnetic radiation, sound, light, algorithm running time, etc., the side channel attack has a huge advantage over the traditional cryptographic attack method . [0003] Since the timing attack proposed by Kocher et al., side channel attack (SCA) has shown great potential in breaking cryptosystems. Side channel attack is a new type of cryptographic attack method. Different from traditional cryptanalysis techniques, side channel attack is n...

AI Technical Summary

Problems solved by technology

The idea of ​​using deep learning for modeling attacks is to model by replacing the probability density estimation problem with a supervised classification problem. However, the two problems are not equivalent, which makes deep learning There are still several major issues

[0006] The first problem is that the indicators used in the field of deep learning to measure the classification effect of the model, such as accuracy (Acc), cannot measure the effect of the side channel attack of the model.

A model with a higher accuracy is not necessarily more effective than a model with lower accuracy, and even using a dedicated validation set to calculate the accuracy cannot solve this problem, because it is determined by the classification problem and the probability Differences in the density estimation problem result from

This problem is especially serious when the labels of the data set become unbalanced, the prediction accuracy of the model will become higher but the attack effect is not as satisfactory

On the other hand, the general side-channel attack evaluation index GE (Guessing Entropy, guessing entropy) / SR (SuccessRate, success rate) lacks advantages in side-channel attacks based on deep learning. The computational complexity of GE / SR and the ability of attackers (The number of collected side channel information) is demanding, requires real attacks and is difficult to embed in deep learning frameworks

At present, there is no index that takes into account efficiency and can accurately reflect the effect of model attacks

[0007] The second problem is that when using deep learning technology for modeling, the training effect of the model is easily affected by data imbalance.

Some schemes choose to directly use the output of the S box instead of its Hamming weight as the label, but it will increase the number of labels and make the training process more complicated; some schemes use resampling technology to balance the data set, but the technical Generally more complicated, the data set needs to be preprocessed and the effect is difficult to predict

Images