May 19, 2023

Today Claudio presented our recent work on Explainable AI (XAI) at IEEE INFOCOM.

We show how to relate raw explanations of state-of-the art XAI techniques to a physical meaning, i.e., which base station is more influential at time of making traffic forecasts with spatio-temporal inputs. Next, we show that harnessing such knowledge fools the predictor in a fully different way than state-of-the art perturbation attacks.

We release our artifacts publicly:

Categorised in: