I was going through the paper (https://link.springer.com/chapter/10.1007/978-3-319-66787-4_3), and I came across where authors says that accuracy alone is not enough as an evaluation metric and guessing entropy should be taken into account as well (Section 2.4). I do understand how accuracy is measured in machine learning but I am unable to understand how to measure guessing entropy for a particular key byte or key bit. How guessing entropy is calculated here, which finally leads to the minimal number of traces? What does 10 independent attacks mean? Does this mean that attack was launched on test data 10 times and mean of all the outcomes was taken?

To take this remark into account, we will always associate the test accuracy to a side-channel metric defined as the minimal number N⋆ of side-channel traces that makes the guessing entropy (the average rank of the right key candidate) be permanently equal to 1 (see e.g. Table 1). We will estimate such a guessing entropy through 10 independent attacks.

  • $\begingroup$ Free version here. $\endgroup$ – Paul Uszak Jul 30 '19 at 11:50
  • $\begingroup$ They use HMMs, which is the same way that you do audio classification. They are looking at the power in the same way as an audio spectrum. The "key word" correlations are the power profile. A minimum number of 10 for the model is surprisingly low. I think there must be more work somewhere as it wasn't clear under a 0th order analysis. $\endgroup$ – b degnan Jul 31 '19 at 9:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.