Abstract
The popularisation of neural networks has seen incredible advances in patternrecognition, driven by the supervised learning of human annotations. However,this approach is unsustainable in relation to the dramatically increasing sizeof real-world datasets. This has led to a resurgence in self-supervisedlearning, a paradigm whereby the model generates its own supervisory signalfrom the data. Here we propose a hybrid quantum-classical neural networkarchitecture for contrastive self-supervised learning and test itseffectiveness in proof-of-principle experiments. Interestingly, we observe anumerical advantage for the learning of visual representations usingsmall-scale quantum neural networks over equivalently structured classicalnetworks, even when the quantum circuits are sampled with only 100 shots.Furthermore, we apply our best quantum model to classify unseen images on theibmq_paris quantum computer and find that current noisy devices can alreadyachieve equal accuracy to the equivalent classical model on downstream tasks.