Explaining Robustness And deriving Robust Explanations – ERARE

The project ERARE aims to develop novel approaches to generate robust explanations in constrained environments (e.g. frugal AI, contexts with limited access to resources or contexts where data is limited or flawed with incompleteness or impreciseness). It also aims to go beyond explaining model decisions to explaining also its predictive uncertainty and its robustness in distribution shift contexts. The outputs of the project will fill a real gap by addressing robustness in XAI and it is complementary with other desirable properties such as expressiveness, intelligibility and consistency. The academic partners composing the ERARE consortium (CRIL, Heudiasyc and MICS ) are public academic laboratories specialized in the project topics and they have complementary skills. The industrial partner (KAPSDATA) is IT company that develops intelligent solutions for real-time monitoring and optimizing energy consumption of data-centers.