Introduction

Machine learning has established itself as a successful interdisciplinary field which seeks to find patterns in data. Throwing in quantum computing gives rise to interesting areas of research that aim to use the principles of quantum mechanics to augment machine learning, or vice-versa. In this page, we aim to give you a glimpse into the exciting and rapidly changing field of near-term quantum machine learning.

Machine Learning (ML)

Before we dive into quantum machine learning, let's do a whirlwind overview of machine learning. For our purposes, machine learning can be split roughly into three subfields: supervised learning, unsupervised learning, and reinforcement learning.

Fields of quantum machine learning

Supervised Learning

Given tuples of labeled data $(x_i,y_i)$ , we aim to learn the function that maps $f: x \mapsto y$ and generalizes to unseen inputs. For example, given a set of labeled photos of cats or dogs, we want to identify new photos of cats or dogs.

Unsupervised Learning

Given a collection of unlabeled data $(x_i)$ , we aim to learn some structure of the data; for example, grouping a set of viewers based their movie viewing history in order to recommend new movies.

Reinforcement Learning

Given access to an environment that rewards us based on our actions, we aim to maximise our expected rewards; for example, algorithmically learning how to play PAC-MAN.

Quantum Machine Learning (QML)

There are four different approaches to combining quantum computing and machine learning, differentiated by whether the data is classical (C) or quantum (Q), or whether the algorithm runs on a classical (C) or quantum (Q) computer. This is illustrated below from Reference 1.

In this context, a quantum dataset consists of observations from a natural or artificial quantum system, such as measurements of qubit interactions, while a classical dataset consists of observations from a classical system, such as times series, text, or images.

QML Approaches

Type of algorithm

Type of data

CC

CC refers to processing Classical data using Classical computers, but using algorithms inspired by quantum computing, such as this recommendation system algorithm.

CQ

CQ refers to processing Classical data using Quantum machine learning algorithms, and will be what this chapter focuses on.

QC

QC refers to processing Quantum data using Classical machine learning algorithms. This is an active area of investigation, with classical machine learning algorithms used in many areas in the quantum computing, such as qubit characterization, control and readout.

QQ

QQ refers to processing Quantum data using Quantum machine learning algorithms. This is an interesting topic, but very much still in its infancy.

Quick quiz

In this course, we'll focus on processing...

Classical data using Classical algorithms

Classical data using Quantum algorithms

Quantum data using Classical algorithms

Quantum data using Quantum algorithms

There are two distinct categories of QC algorithms: those that require quantum random access memory (qRAM), where data can be accessed in superposition, and those that don't. The various proposed qRAM-based QML algorithms, e.g. qPCA, qSVM and qClustering, boast exponential speedups compared to their classical algorithms, however there are currently no viable hardware candidates for realizing qRAM.

Recently, most of the focus of CQ approaches to machine learning have been on near-term algorithms that can be executed on the current quantum devices. Classical machine learning techniques have made great strides in the past decade, enabled in large part by the availability of sufficiently powerful hardware. Maybe the existence of quantum hardware might enable further advances in the field.

Note that this is a very dynamic area of current investigation by multiple research teams worldwide, and there are still many open questions, as well as confusing terminology and notation. We will try and be as consistent as possible in this course, but also point out when concepts have been referred to by different names, or how different concepts relate to each other.

References

Vedran Dunjko, Jacob M. Taylor and Hans J. Briegel, Quantum-Enhanced Machine Learning, Physical Review Letters 117 (13), 130501 (2016) doi:10.1103/PhysRevLett.117.130501 arXiv:1610.08251