2020 IEEE Third International Conference on Artificial Intelligence and Knowledge Engineering (AIKE)

Overview on Quantum Computing and its

Applications in Artificial Intelligence
Nahed Abdelgaber Dr. Chris Nikolopoulos
Computer Science and Information Systems Computer Science and Information Systems
Bradley University Bradley University
Peoria, Illinois, USA Peoria, Illinois, USA
Email: [email protected] Email: [email protected]

Abstract—This paper will review the basic building blocks 1.1 T HE Q UANTUM BIT
of quantum computing and discuss the main applications in
artificial intelligence that can be addressed more efficiently using
The Quantum bit (qubit) is the most basic unit of operation
the quantum computers of today. Artificial intelligence and in quantum computing. It is also represented as a 2 dimen-
quantum computing have many features in common. Quantum sional vector space over the complex numbers. The standard
computing can provide artificial intelligence and machine learn- |0> and |1>. 
basis are 
ing algorithms with speed of training and computational power in 1 0
less price. Artificial intelligence on the other hand can provide |0> = |1> =
0 1
quantum computers with the necessary error correction algo-
rithms. Some of the algorithms in AI that have been successfully
A qubit is not always in either |0> or |1>, it can be an
implemented on a quantum computer, that we will present in arbitrary quantum state |ψ>, which can be any superposition
this paper, are both unsupervised learning algorithms (clustering |ψ> = α|0> + β|1>. where α and β are complex numbers;
and Principal component analysis) and also supervised learning α2 + β 2 =1.
classification, such as support vector machines. Since qubits are 1 and 0 concurrently, with a certain
probability of being a 1 and a 0, they can be viewed as being
INTRODUCTION in many parallel universes. For example, if we have a qubit
and it split with probabilities to 2 parallel universes, it can be
The interest in using quantum computers to compute arti- combined again as a photon with either 1 or 0. So, this allows
ficial intelligent algorithms has increased exponentially with faster calculations in different parallel universes.
the many successful implementations of AI showing that we 1.2 S UPERPOSITIONS
can rely on quantum computers for calculations. In section
1 of this paper, we will discuss the fundamentals of quantum To go beyond the digital behavior, we can use the Hadamard
computing from the perspective of the programming paradigm gate(H) to transform the qubit to the superposition state. Any
of quantum computing. linear combination of two quantum states, once normalized,
will also be a valid quantum state. The matrix representation
In section 2, we discuss a paper ([1]) where the authors
of the Hadamard
 gate  is:
examine applications of quantum computers and how it will √ 1 1
affect the humanities. They also showed the advantages of H = 1/ 2
1 −1
using quantum computers over the classical computers as they For example
 applying
   the H tothe  |0> state will be H|0>
are faster and more accurate, and they can solve problems that √ 1 1 1 √ 1
classical computers cannot. = 1/ 2 =1/ 2
1 −1 0 1
In section 3, we discuss machine learning in quantum
spaces ([4]). Schuld examined the interplay between machine 1.3 Q UBIT GATES
learning and quantum computing and explains that the state of A qubit “gate” performs operations on qubits. Each gate
a quantum system cannot be simulated in a classical computer can be considered as a matrix multiplication filter. Some gates
because of the exponential number of bits needed to just which act on one single qubit:
simulate an intermediate quantum step. Pauli-X gate: like the digital NOT gate, it will flip the qubit
from 1 to 0 and vice versa. It maps 1> to 0> and 0> to 1>.
0 1
1. Q UANTUM COMPUTING BASIC BUILDING BLOCKS X=
1 0
Pauli-Y gate: It rotates the qubit around the Y-axis of the
In digital computers all information is reduced to a sequence Bloch sphere by π radians. It maps 0> to i|1> and 1> to
of bits and all computations are done through logic gates. The i|0>.  
digital computer’s state is determined by the state of its bits, so 0 −i
a computer with n states can exist in one of 2n possible states. Y =
i 0
A quantum computer also has bits, but instead of 1’s and 0’s, Pauli-Z gate:
the quantum bit can represent both a linear combination of 1 It rotates the qubit around the Z-axis of the Bloch sphere
and 0 at the same time, and this is known as superposition. by π radians.

978-1-7281-8708-2/20/$31.00 ©2020 IEEE 198

DOI 10.1109/AIKE48582.2020.00038
  
1 0 3.1.2 A LGORITHM DESCRIPTION
Z=
0 −1 A density matrix defines probabilistic mixtures for quantum
states. Let’s assume that we have a quantum system whose
1.4 E NTANGLEMENT
state is not known, rather we know that it can be in one of M
The state of any n qubit system can be written as a nor- states, |ψ>, with probability pi . The state of
the system repre-
malized linear combination of the 2n bit-string states (states M
sented by density matrix ρ defined as: ρ = i=1 pi |ψ> < ψ|
formed by the tensor products of |0>’s and |1>’s. It is possible If the state of a system is known with probability 1, then
for three qubits to be in a state that cannot be written as the the system in a pure quantum state and the density matrix will
√ product of three single qubit states. For instance, |ψ> =
tensor only be |ψ> < ψ| otherwise the system in a mixed state.
1/ 2(|000> + |111>). Entanglement provides more power to Any mixed state system of n qubits can be part of a larger
quantum computing by making it possible to create a complete system which is in a pure state. For instance, a mixed state n
2n dimensional complex vector space for computations, using qubit system can be purified to a pure state system by adding
just n physical qubits. more n qubits, ending up with 2n qubit system in a pure state.
Back to PCA  Algorithm, It can be done in 4 steps:
1.6 M EASUREMENTS 1- encode in a quantum density matrix ρ
Measurements is converting quantum information into clas- 2- arrange many copies of ρ
sical bits. A significant principle of quantum mechanics is that 3-perform the exponential SWAP operation on each copy
measurement outcomes are probabilistic. and a target system
In single qubit the possibility of obtaining |0> after mea- 4-perform quantum phase estimation to determine the eigen-
surement is | < 0|φ>|2 , and the possibility of obtaining |1> values
after measurement is | < 1|φ>|2 . 3.2 QUANTUM SUPPORT VECTOR MACHINE
the probability of obtaining n string of bits |x1 x2 ...xn is
3.2.1 P ROBLEM DEFINITION
| < x1 x2 ...xn |φ>|2 .
SVM is a supervised ML algorithm for binary classification.
2. C LUSTERING PROBLEM IN QUANTUM COMPUTING Having M data points Xj , yj as j = 1, 2, ..., M and yj is the
Algorithms for clustering are unsupervised machine learn- label. SVM solves the equation W X + b = 0.
Least square SVM is a version of SVM it approximates the
ing algorithms. In ([2]), a hybrid environment of classical
and quantum computation was introduced. The vectors to be  plane by
hyper   the
solving  following
  linear equation
0 1T b 0
clustered are in k feature space and the distance between =
1 K + γ −1 1 α Y
each two elements is the Euclidean distance. The clustering
problem can be reformed as a fully connected undirected 3.2.2 A LGORITHM DESCRIPTION
weighted graph. The weight between two elements is the Input: 1- Training dataset Xj , yj as j = 1, 2, ..., M 2- a
distance between them, thus the weighted maximum cut of query data X Output: classification of X: +1 or -1
the graph will give us separate groups of the data points. The Algorithm: 1- calculate kernel matrix K = Xj Xi 2- solve
weighted maximum cut of a graph can be found using the linear equation and find |b, α> 3- perform the query result
quantum approximation optimization algorithm (QAOA). In by using the |b, α> in the linear equation. The quantum
section 3.2 we train a quantum SVM, so that any new data computing complexity for least square SVM is O(log N M ).
element is predicted as belonging to one of the pre-identified CONCLUSION
clusters.
In this paper, we have reviewed the main building blocks of
3. Q UANTUM COMPUTING ALGORITHMS IN ARTIFICIAL quantum computing and investigated some of the applications
INTELLIGENCE in the field of artificial intelligence. The applications have
showed a great impact on both supervised machine learning
In this section, we implement and examine two quantum models like support vector machines and also on unsupervised
computing algorithms from the field of AI, namely the quan- learning like using clustering techniques and principal compo-
tum PCA algorithm and the quantum SVM. nent analysis. The results support the conclusion that the use
3.1 QUANTUM PRINCIPAL COMPONENT ANALYSIS of quantum algorithms in artificial intelligence techniques can
boost machines’ learning abilities and that the use of quantum
3.1.1 P ROBLEM DEFINITION computing on applying machine learning can speed up the
PCA is a dimensionality reduction algorithm used in ML to training time of the various algorithms.
reduce the number of features. It does this by projecting the R EFERENCES
input instance vectors along the best k eigenvectors. The goal
[1] Barzen, J., Leymann, F. (2019). Quantum humanities: a vision for quan-
of PCA is to pick the k most important features that capture tum computing in digital humanities. SICS Software-Intensive Cyber-
the most variance in the data. PCA calculates the covariance Physical Systems, 1-6.
matrix for [2] Coles, P. J., Eidenbenz, S., Pakin, S., Adedoyin, A., Ambrosiano, J.,
 two features
 as in this equation:
 e1 0 Anisimov, P., ... Gunter, D. (2018). Quantum algorithm implementations
= for beginners. arXiv preprint arXiv:1804.03719.
0 e2 [3] Crooks, G. E. (2018). Performance of the quantum approximate op-
From the diagonal values of the covariance matrix we can timization algorithm on the maximum cut problem. arXiv preprint
chose the k most effective features, and k will be a free arXiv:1811.08419.
[4] Schuld, M. (2019). Machine learning in quantum spaces.
parameter.

199