What Can We Expect from Quantum Machine Learning?

Dr. Peter Wittek
Spain, ICFO-The Institute of Photonic Sciences, Barcelona
The most successful algorithms in machine learning strike a balance between sample, model, and computational complexities. A low sample complexity means that the algorithm can work with only a little training data. Model complexity will influence the generalization performance, telling whether previously unseen data instances will be recognized correctly. A lower computational complexity will allow the algorithm to scale to larger amounts of data. Over the last two decades, several proposals have been put forward to perform machine learning using a quantum system. Advantages range from quadratic or even exponential speedup, reduced sample complexity and better generalization performance of the learned models. Examples include optimization through adiabatic annealing, quantum support vector machines based on a quantum random access memory, neural networks implemented with quantum dots or quantum optical systems, and quantum deep learning using state preparation and sampling. Some of these even have experimental demonstrations using actual quantum hardware. In this talk, we give an introduction to the most relevant concepts in quantum mechanics and quantum information theory to understand the current practical trade-offs. Then we overview the major research directions in quantum machine learning and discuss the feasibility of some of the proposed models.