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… Read More »Estimation of uncertainty
Research When a neural network model has been trained on a set of images representing the realizations of the input data distribution, it performs well on test data from the same distribution. However, in a large number of applications, the… Read More »Domain adaptation
Research Most current network models are trained by supervised bases. The cost of building these databases is high. Active learning consists in using a partially trained model to select the next set of data to annotate. These techniques allow to… Read More »Active learning