DQC Seminar Series: Quantum Machine Learning via Contrastive Training

Abstract: How can quantum hardware learn useful features when labeled data are scarce? In this talk, I will present our recent work demonstrating quantum machine learning via contrastive training on a programmable trapped-ion quantum computer. We use a self-supervised approach to pretrain quantum representations from unlabeled examples, with image similarity defined directly through measured quantum overlaps on hardware. This allows both representation learning and classification to be performed in situ on the quantum processor. We find that contrastive pretraining leads to higher classification accuracy, lower variability, and especially strong improvements when only limited labeled data are available. Beyond the specific image-classification setting, this work suggests a practical path toward label-efficient quantum representation learning on near-term devices. I will conclude by discussing how this framework could be extended to larger images, richer datasets, and physics-driven applications, including settings motivated by condensed matter and many-body quantum systems. --- Upcoming seminars 2 Apr: TBA 16 Apr: TBA 23 Apr: Sophia Economou