Research
To estimate the uncertainty of the output of a network consists in inferring several times the same input by randomly deleting a part of the network connections. The statistics of the output decisions translate a probability function on the different decisions, thus encoding the associated uncertainty. We work in collaboration with the Computer Science Laboratory of Clermont-Ferrand (LIMOS) in the context of the detection of Out-Of-Domain data using belief theory. For more information, you can consult our paper published in the International Joint Conference on Neural Networks, 2021 or the project: https://github.com/Soplia/SLUE-demo