cassiersg
commented
2 years ago Triage report:
The chi-squared test seems indeed as useful addition, however we don't see the use for -log10(p) > 300 as this level of confidence seems to be sufficient for any use-case.
It might also be worth considering the G-test instead or in addition to the T-test (G-test might be more accurate, so let's use that if it is not computationally more expensive).
Hi,
I would like to ask if you could add support for leakage detection with the chi-squared test to scalib. The C++ implementation provided in [1] which uses the boost library seems to be very slow when a high accuracy is needed (e.g. -log10(p) values > 300). An efficient option to produce -log10(p) values using the t-test would be great for comparability, too (as e.g. done in [1] and [2]).
[1] https://tches.iacr.org/index.php/TCHES/article/view/838/790 [2] https://tches.iacr.org/index.php/TCHES/article/view/8986/8563