With quantum computing technologies nearing the era of commercialization andquantum supremacy, machine learning (ML) appears as one of the promising"killer" applications. Despite significant effort, there has been a disconnectbetween most quantum machine learning proposals, the needs of ML practitioners,and the capabilities of near-term quantum devices to demonstrate quantumenhancement in the near future. In this contribution to the focus collection on"What would you do with 1000 qubits?", we provide concrete examples ofintractable ML tasks that could be enhanced with near-term devices. We arguethat to reach this target, the focus should be on areas where ML researchersare still struggling, such as generative models in unsupervised orsemisupervised learning, instead of the popular and much more tractable MLtechniques. We also highlight the case of classical datasets with potentialquantum-like statistical correlations where quantum models could be moresuitable. We focus on hybrid quantum-classical approaches and illustrate someof the key challenges we foresee for near-term implementations. Finally, weintroduce the quantum-assisted Helmholtz machine (QAHM); an attempt to usenear-term quantum devices to tackle high-resolution datasets on continuousvariables. Instead of using quantum computers to assist deep learning, asprevious approaches do, the QAHM uses deep learning to extract alow-dimensional binary representation of data, suitable for relatively smallquantum processors which can assist the training of an unsupervised generativemodel. Although we illustrate this concept on a quantum annealer, other quantumplatforms could benefit as well from this hybrid quantum-classical framework.
展开▼
