Figure 5
From: Quantum adversarial metric learning model based on triplet loss function
The inner products between test sample pairs of MNIST. The horizontal and vertical axes represent the indexes of samples. Indexes 0-49 denote the samples from class ‘0’, and indexes 50-99 represent the samples from class ‘1’. Panel (a) shows the inner products between all sample pairs before performing the QML model, also corresponding to the inner products before performing the QAML model. Panel (b) shows the inner products of test sample pairs, where the QML model is trained through 1000 training epochs but adversarial samples are not added to the training set. Panel (c) shows the inner products of test sample pairs after 1000 training epochs, where adversarial samples are added to the training set