Uncertainty of feed forward neural networks recognizing quantum contextuality

Abstract

The usual figure of merit characterizing the performance of neural networks applied to problems in the quantum domain is their accuracy, being the probability of a correct answer on a previously unseen input. Here we append this parameter with the uncertainty of the prediction, characterizing the degree of confidence in the answer. A powerful technique for estimating uncertainty is provided by Bayesian neural networks (BNNs). We first give simple illustrative examples of advantages brought forward by BNNs, out of which we wish to highlight their ability of reliable uncertainty estimation even after training with biased datasets. Then we apply BNNs to the problem of recognition of quantum contextuality, which shows that the uncertainty itself is an independent parameter identifying the chance of misclassification of contextuality.