no code implementations • 10 Sep 2023 • Pavel Burnyshev, Elizaveta Kostenok, Alexey Zaytsev
Through our investigation, we provide evidence that machine translation models display robustness displayed robustness against best performed known adversarial attacks, as the degree of perturbation in the output is directly proportional to the perturbation in the input.
no code implementations • 22 Aug 2023 • Elizaveta Kostenok, Daniil Cherniavskii, Alexey Zaytsev
In this paper, we propose a method for uncertainty estimation based on the topological properties of the attention mechanism and compare it with classical methods.