1

Classification with Quantum Machine Learning:

A Survey
Zainab Abohashima 1 | Mohamed Elhoseny 2 | Essam H. Houssein 3 | Waleed M. Mohamed 3


1
Faculty of Computer Science, Nahda University, Egypt
2
Faculty of Computers and Information, Mansoura University, Egypt
3
Faculty of Computers and Information, Minia University, Egypt

Abstract
Due to the superiority and noteworthy progress of Quantum Computing (QC) in a lot of applications such as cryptography,
chemistry, Big data, machine learning, optimization, Internet of Things (IoT), Blockchain, communication, and many more.
Fully towards to combine classical machine learning (ML) with Quantum Information Processing (QIP) to build a new field in
the quantum world is called Quantum Machine Learning (QML) to solve and improve problems that displayed in classical
machine learning (e.g. time and energy consumption, kernel estimation). The aim of this paper presents and summarizes a
comprehensive survey of the state-of-the-art advances in Quantum Machine Learning (QML). Especially, recent QML
classification works. Also, we cover about 30 publications that are published lately in Quantum Machine Learning (QML). we
propose a classification scheme in the quantum world and discuss encoding methods for mapping classical data to quantum
data. Then, we provide quantum subroutines and some methods of Quantum Computing (QC) in improving performance and
speed up of classical Machine Learning (ML). And also some of QML applications in various fields, challenges, and future vision
will be presented.

KEYWORDS
Quantum Machine Learning, Quantum Computing, Quantum Bit (Qubit), Quantum Inspired, Hybrid Quantum-Classical,
Variational Quantum Classifier, Quantum Classification, Machine Learning.

1 | INTRODUCTION

As is well known the role of machine learning (1-4) in data analysis , feature selection, making decision, pattern
classification and future predictions of various applications with achieving better accuracy and performance without
human decisions. but an enormous increase in types of data (i.e., images, text, videos, and record audios) and current
computational resources appeared problems of machine learning such as high-cost learning and kernel estimation.
Over the last three decades, we noticed using quantum computing (QC) in various applications as cryptography (5),
artificial intelligence (AI) (6), and communication (7). As a result of the superiority and advancement of quantum
computing for solving many problems as factorization by Shor’s algorithm (8), and search in an unstructured database
by Grover’s algorithm (9).
Quantum computing based on postulates and characteristics of quantum mechanics (i.e., quantum bits (Qubits),
interference, superposition, and entanglement) to information processing. qubit can be one state, zero state, or a
combination of two states at the same time known as linear superposition, unlike classical bit that can represent one
value either 0 or 1 to store information (7). Qubit state is a unit vector in Hilbert space. mathematically to represent
qubit state, we use ket- notation, qubit in state zero is |0⟩= [1 0]T and qubit in state one is |1⟩ ⟩= [0 1]T . a qubit
is represented as a linear superposition of both basis states simultaneously:
 2

|ψ⟩= (α𝛽 )= α|0⟩ + 𝛽|1⟩ (1)

Where the coefficients α and 𝛽 are probability amplitudes may be complex numbers and | α2 | + | 𝛽2 | = 1.
Recently, different algorithms and approaches have been published in machine learning (ML) depends on quantum
computing laws to introduce the new concept called quantum machine learning (QML) )4( (10, 11) with goal
improvement of classical ML (12). QML appeared in more than a paradigm. the first paradigm, quantum ML, in 2014
Rebentrost, p et al )24( presented a quantum version of support vector machine (QSVM) for classification of big data.
QSVM achieved logarithmic speed up over classical counterparts. The second paradigm, Quantum inspired ML, in
2019 Sergioli et al. )28( proposed a binary quantum classifier inspired by the formalism of quantum theory and this
classifier achieved a higher performance compared with different classical models. In 2019 Havlicek et al. )40(
presented a quantum-classical classifier depends on the concept of the variational quantum circuit and this paradigm
named hybrid quantum-classical ML. Also, various models have developed in Quantum Neural Networks (QNNs) such
as QNNs )25(, Quantum inspired NNs )37(, Quantum multi perceptron NNs, quantum dot NNs, and quantum
convolutional NNs. in detail, all these paradigms will be presented and comparison among them in section 2. QML is
used in several applications as medical data classification )30,43(, recommendation systems, and big data processing
)24(.
Quantum subroutines are the lifeblood of quantum and QML algorithms. Quantum subroutines have a major role in
speed up and improve the performance of algorithms such as HHL algorithms )88( for solving “linear system
equations” in exponential speed up based on quantum phase estimation (QPE) and quantum matrix- inversion
subroutines and Grover’s algorithm based on amplitude amplification. Other subroutines as sampling, quantum
annealing, and quantum fourier transform (QFT).
The main objectives of this survey introduce a classification scheme with QML. To implement a classical algorithm on
a quantum computer or in general any quantum algorithm consists of the main three phases (13, 14): encoding,
quantum computation, and decoding phase. first of all, the encoding phase that means mapping data from classical
to quantum states. In the second step, quantum computation depends on type QML algorithms. the last step,
decoding phase. In addition to illustrating the major phases of quantum classification problems to answer the
question of survey what is the classification scheme with quantum machine learning?
We will open a discussion about two questions in QML algorithms. Why do we proceed towards Quantum Machine
learning (QML)? How does the concept of quantum computing enhance classical machine learning? The main
objectives of this survey are:
1. Summarize and organize the most recent research works to pave the way for researchers in quantum
machine learning.
2. Analyze techniques and show the best and most used methods in classification real problems.
3. Provide the readers with various quantum methods to enhance classical ML and some of the quantum
subroutines.
4. Introduce quantum classification scheme in the quantum world.
5. Present some of the challenges, future directions, and applications of quantum machine learning

The remainder of the survey can be divided into six sections: section 2 presents the major and recent approaches to
quantum machine learning techniques especially quantum classification algorithms classified into 3 categories QML,
quantum-inspired ML, and hybrid quantum-classical ML. Also, summarizes these techniques in table 1. Next, we
present the concept of classification with QML, two methods of encoding data, variational quantum circuits, and,
proposed quantum classification scheme in section 3. Thereafter, a discussion of the survey questions in section 4.
section 5 introduces some of the challenges and future directions of QML. We provide many applications of QML in
section 6 Finally, concludes the survey in section 7

2 | LITERATURE REVIEW

Here, we present exhaustive previous work about QML. As well as, we classify quantum machine learning algorithms
 3

into three approaches: QML, quantum-inspired ML, and Hybrid quantum-classical ML and summarized in table 1 &
table 2. The classification of approaches based on information processing devices (a type of algorithm) is classical or
quantum and type of data are classical or quantum (15, 16).
Montanaro, A. (17) introduced an overview of quantum algorithms and some of the applications in different areas.
Jeswal, S. et al. (18) presented various quantum neural network techniques and their applications in real-world
problems. the authors discussed each technique in detail and noted that QNNs are more powerful and enhance
computational efficiency in contrast to classical NNs. Benedetti, M et al. (19) offered an overview of hybrid quantum-
classical models based on parameterized quantum circuits and the application of hybrid systems in supervised and
generative learning. They also provided a framework for components of models as an example, variational circuit and
encoder circuit. In (20), the authors given an overview of the progress of quantum computing and applications. Also,
the authors discussed various quantum technologies for quantum computers scalability such as error correction and
going forward. Ciliberto, C et al.(21) presented a review about QML and challenges. AS well, the authors provided
some quantum subroutines especially quantum linear algebra and how a quantum computer works with data.
Besides, this review illustrates Quantum neural networks.

A. Quantum machine learning algorithms

The first approach, quantum machine learning algorithms are quantum versions from conventional ML. as well,
algorithms that can be executed on the real quantum device. Dennis, et al. (22) implemented SVM on quantum
annealer device (23) (DW2000Q) called QA-SVM. The authors used quantum annealer to train and optimize SVM
depends on QUBO equation to minimize cost energy. As well, the authors utilized some of feature of quantum
annealing (i.e. reverse annealing, and special annealing schedules to improve final results. Rebentrost, P. et al (24)
presented SVM algorithm runs quantum computer on and depends on a non-sparse matrix called QSVM. QSVM is a
big data binary classifier. As well, it works with a large number of features and samples in complexity logarithmic. da
Silva, A. et al (25) introduced a new quantum neural network named “quantum perceptron over a field” (QPF) and
its learning algorithm (SAL). the learning algorithm (SAL) based on superposition feature and quantum operator. Also,
it performs NN architecture with polynomial time. QPF overcomes on limitations of quantum perceptron models. In
(26) the authors proposed a version of linear regression called quantum linear regression.it works on quantum data
with N- dimensions of features in logarithmic time.

B. Quantum-inspired machine learning

The second approach, quantum-inspired machine learning that applies the principles of Quantum Computing (QC) to
improve classical methods of machine learning (ML).
Prayag et al. (27) introduced a new quantum-inspired binary classifier (QIBC), the basic idea is based on decision
theory, classical ML and theory of quantum detection that utilize one of the laws of quantum mechanics,
superposition to increase the higher degree of freedom in decision making. The proposed classifier can be achieved
high precision, recall and F-measure comparable with KNN, and SVM and other classical techniques. Sergioli et al.
(28) proposed a novel quantum-inspired classifier for binary supervised learning called Helstrom Quantum Centroid
(HQC) based on density matrices and formalism of quantum theory. The authors evaluated the performance of their
model by fourteen datasets compared to different classical models. Ding et al. (29) proposed a novel algorithm
inspired by the quantum support vector machines(SVM) to solve classification problems in exponential speedup. The
main idea for the algorithm based on linear transformation.
Sergioli et al. (30) introduced a Quantum Nearest Mean Classifier (QNMC) based on the idea of classical minimum
distance classifier. The algorithm consists of three main steps: firstly, density pattern (Encoding) to transform each
classical data point to a quantum object. Secondly, Quantum centroid to calculate a distance among density patterns
in order to classify unknown quantum objects to the right class. lastly, decoding to transform the final classification
result into the classified data. The algorithm achieved higher accuracy in many medical data sets than the classical
counterpart (NMC) exclusively cancer data set. Dang, Y et al. (31) proposed a new model based on Quantum KNN
 4

and parallel computing for image classification. their model improved efficiency and classification performance. Chen,
H. et al. (32) used powerful parallel computing to introduce an inspired Quantum K Nearest-Neighbor (QKNN) based
on one of the well-known properties of QC is superposition to obtain parallel computing and “quantum minimum
search algorithm” to achieve speed up the search.
Lu, S et al (33) proposed a quantum version for decision tree classifier. The quantum model depends on quantum
entropy impurity and quantum fidelity measure. In (34) the authors developed a new model Quantum Support
Vector Clustering for big data depended on Quantum SVM, quantum Gaussian kernel and quantum polynomial
kernel. In (35), The authors proposed new framework quantum clustering based on the Schrödinger equation. Yu,
C.et al. (36) presented improved quantum techniques for ridge regression. This algorithm based on two subroutines
quantum K-fold cross-validation and quantum state encoding. In (37) the authors introduced a novel Quantum
inspired Neural network. they named Autonomous Perceptron Model (APM) depends on one from quantum
mechanics characteristics a quantum bit. The proposed model accomplished higher accuracy with less complexity
time comparable to other tradition-al algorithms.

C. Hybrid quantum-classical machine learning

The last approach, hybrid quantum-classical machine learning are the algorithms that combine quantum algorithms
and classical (traditional algorithms) to obtain higher performance and decrease in the learning cost.
Soumik et al. (38) used the quantum circuit to present a new variational quantum classifier with a single quantum
system (Qu N it) to encode data in N-dimensional with a training algorithm called “single-shot training”. the key
advantage of single-shot training uses a fewer parameter for training and achieve a higher precision. In (39), the
authors introduced a new quantum algorithms based on many subroutines as a quantum oracle, counting, amplitude
amplification, and quantum amplitude estimation for feature selection named (HQFSA) with purpose enhancement
of performance ML techniques. The proposed algorithm accomplished quadratic time complexity and better
performance in some of the cases. The main disadvantage of HQFSA is running on a quantum simulator only.
Havlicek et al. (40) suggested two different models of quantum support vector machines. the first, the variational
quantum SVM based on quantum variation circuit. The variational quantum SVM classifier requires two algorithms
to classify. The one to quantum variational training phase, in this phase, the authors used four steps to compute the
hyperplane between training data and another to quantum variational classification phase to classify new datum to
correct label. The second model is a SVM quantum kernel-based algorithm based on quantum kernel estimation.
Maria Schuld et al. (41) proposed two- hybrid quantum techniques for classification problems. Schuld showed that
quantum computing enhances classical ML algorithms like kernel methods. Quantum computing performs complex
computations in Hilbert space more efficient. The authors focused on using feature maps and kernel methods in the
quantum computing world.

Mitarai, K et al (42) presented a hybrid quantum-classical technique to perform different tasks like classification,
regression, and clustering to can implement it on the small-scale quantum devices. Jessica Pointing introduced (43)
in the Ph.D. thesis a novel quantum-classical algorithm to handing missing values in data is named “a Quantum
Algorithm for Handling Missing Data”. The main advantage of the algorithm calculates the probability distribution to
handle missing data values.

Ruan, Y et al. (44) presented a quantum KNN algorithm based on the hamming distance matrix. QKNN is a good analog
for the classical KNN algorithm, which avoids the defect of the simplified assumption of the task of classification
QKNN outperforms Centroid and QNN on time performance and classification accuracy. Grant, E et al. (45) introduced
a new hybrid classifier based on a hierarchical structure for quantum circuits for binary classification problems. Zhang,
D. B et al (46) presented a quantum version of nonlinear regression rely on the hybrid quantum device. also, the
authors proposed a new encoding method. Benedetti, M. (47) introduced a framework generative model named
“data-driven quantum circuit learning” (DDQCL). DDQCL approach is unsupervised hybrid quantum- classical for the
characterization of NISQ hardware for solving sampling problems such as bars and stripes (BAS) and random thermal
 5

data sets by using shallow quantum circuits.

Research Model Class Task Application

Soumik et al (38). Single-shot Hybrid Classification Cancer, Sonar, and Iris
training
Sanjay, et al. (39) HQFSA Hybrid Feature Breast Cancer, Iris, Wine, Vehicle,
selection Glass, Sonar, and Ionosphere.
Dennis, et al. (22) QA_SVM Quantum Classification Real and synthetic data
and optimization
Prayag et al. (27) QIBC In-spired Classification Text and image Corpora data set

Sergioli et al. (28) HQC In-spired Classification -

da Silva, A. et al (25) QPF Quantum Classification -

Ding et al. (29) QiSVM In-spired Classification -

Casaña-Eslava, R. Probabilistic In-spired Clustering Local densities &

(35) Quantum Two spirals
Clustering (PQC)
Yu, C.(36) - In-spired Regression -

Benedetti, M. (47) DDQCL Hybrid Generative Synthetic data

model
Jessica Pointing (43) - Hybrid Handling missing Heart data set
values
Maria Schuld et al. ( - Hybrid Classification -
41)
Lu, S et al (33) MERA and TTN Hybrid Classification MNIST digit and synthetic data
set
Sergioli et al. (30) QNMC In-spired Classification Medical Data

Dang, Y et al. (31) QKNN In-spired classification Graz-01 and Caltech-10 datasets

Schuld, M et al (48) Circuit-centric Hybrid classification

quantum classifier Cancer and MNIST
Bishwas, A. (34) Binary clustering In-spired Clustering Big Data
QSVM
Zhang, D. B et al (46) Hybrid Regression
Ruan, Y et al (44) QKNN Hybrid Classification MNIST digit data set

Schuld, M et al (26) Quantum linear Quantum Regression -

regression
Chen, H. et al (32) QKNN In-spired Classification -

Rebentrost, P et al. QSVM Quantum Classification big -

(24) data
Lu, S et al (33) Quantum decision In-spired Classification -
tree
Sagheer, A et al (37) APM In-spired Classification Breast cancer and synthetic data

Mitarai, K et al (42) QCL Hybrid Classification & -

Clustering

TABLE 1 Summarizes quantum machine learning algorithms

 6

Algorithm Classification Type Based Idea Q Data Set Advantages Limitations

Data
QIBC Binary Quantum ✔ image and High precision, recall, Need to many features to
detection theory text corpora and F-measure. improve performance.
data set Less computation time Increase computation cost
compared to SVM and with large number copies.
KNN.
HQC Binary distinguishability ✔ Appendicitis Improve classification Don’t work efficiently with
between and other rate. many features vectors in
quantum states data sets large data set.
QNMC Binary & quantum ✔ Diabetes, Achieve higher Don’t achieve better
Multi -class formalism Liver accuracy than classical performance with cancer
Cancer NMC data.
Ionosphere
variational Binary & Quantum ✔ Breast Higher performance High cost
quantum Multi -class Variational Cancer Classy binary & multi-
classifier circuit and wine class efficiently.
data
QSVM Binary & Run on real ✔ Breast Better accuracy from Difficult implementation
Multi -class quantum device Cancer and classical SVM.
wine data
set
APM Binary & Quantum Bit x Breast Classify specific Work on classical computer
Multi -class Cancer nonlinear problems only.
Wine with one neuron only.
Vintage Take a few instances to
synthetic train model and
data achieve higher
accuracy.

TABLE 2
A comparative table among QML classification from through classification type, based idea for algorithm, Q data refers to
algorithm works on quantum data (✔) or not (x), data sets that algorithm use it for evaluation the performance, Advantages
and limitations of algorithm.

3 | QUANTUM CLASSIFICATION SCHEME

Throughout this section, we present the concept of quantum classification, encoding methods, variational quantum
circuits and Finally, we discuss the quantum classification scheme.

a) Classification concept in quantum domain

Classification is a popular task in a supervised learning domain that maps given inputs data (x) to discrete target
output (y) through a function approximation f as follows y = f(x). The main purpose of classification is an accurate
prediction model. Classification problems can be classified into two main problems: classification with two-classes is
known as binary classification (for instance cancer diagnosis and spam detection). Classification with more than two-
classes is known as multi-class classification (for instance image and digit classification). classification problem can be
represented with classical ML domain as C = {𝑐1 , 𝑐2 , … , 𝑐n } where C is target labels and a set of data in training phase
with as 𝐷𝑛 = {(𝑥1 , 𝑦1 ), …, (𝑥i , 𝑦𝑖 ), …, (𝑥n , 𝑦𝑛 )} where xi is some of the features (n) on the properties of order of data
point (i) and 𝑦𝑖 is the corresponding of that data point. In the case of binary classification 𝑦𝑖 ∈ {𝑐1 , 𝑐2 } and xi ∈ 𝑅 𝑑 . in
the case of multi-class classification 𝑦𝑖 ∈ {𝑐1 , ….., 𝑐n } where xi ∈ 𝑅 𝑑 and d is real-valued attributes. To can describe
classification problems with the QML domain, we should convert classical data to quantum data then we can be
represented quantum data in training data as 𝐷𝑛 = {( |𝜓1 ⟩, 𝑦1 ), … , ( |𝜓𝑖 ⟩, 𝑦𝑖 ),…, ( |𝜓𝑛 ⟩, 𝑦𝑛 )} where |𝜓𝑖 ⟩ is the order
𝑑
(i) of the quantum state of 𝐷𝑛 , |𝜓𝑖 ⟩ ∈ ∁2 and 𝑦𝑖 ∈ {c1, c2 }[33],(49) .in the case of binary classification, How to map
data to quantum form? There is more than method (or technique) for encoding classical data to quantum data such
as basis encoding and amplitude encoding for in detail (in next subsection b)- Encoding methods)
 7

b) Encoding methods
There are a lot of encoding methods from classical data to quantum data in a Hilbert space. In other words,
encoding data means loading classical data into quantum computer (quantum states). We outline two
methods. for other methods, we refer the reader to (41) (50) (72).

Basis Encoding is the simplest method to encode data to quantum data. This method associates between n-bit
classical input and the computational basis of n-qubit input. For example, (1100) classical input string is encoded to
four qubits ( |1100⟩) quantum states. In general, to encode data set by basis method, in other word, to represent
data in computational basis states of qubits, we use the following equation.
1
|𝐷⟩ = ∑𝑀𝑚=1 |𝑋 ⟩
𝑚
(2)
√M

Where D = {𝑋1 , 𝑋 2 ,….., 𝑋 𝑀 } is classical data that is in form binary string ,𝑋 𝑚 ={𝑏1 , 𝑏2 , … 𝑏𝑁 }, 𝑏𝑖 ∈ {0,1} , i ∈ {
1,2,….,N} and N is number of features.

Amplitude Encoding is the most used and popular method encoding in QML algorithms is amplitude encoding. The
main idea of amplitude encoding based on association classical data with quantum state amplitudes. To encode
classical data vector to quantum amplitudes. We should convert the classical vector to a normalized classical vector
(28,49).
𝑥1
𝑥2
X= [ : ] (3)
:
𝑥2 𝑛
𝑛
Where X is normalized classical vector , x ∈ Ϲ2 , and is Ϲ complex numbers.
quantum state amplitudes can be encoded as follows:
𝑛
|𝜓𝑥 ⟩ = ∑2𝑖=1−1 𝑥𝑖 |i⟩ (4)

where |𝜓⟩ ∈ Hilbert space(ℋ) and ∑𝑖|𝑥𝑖|2 =1

c) Variational quantum circuit

Variational quantum circuits also known as parameterized quantum circuits, most hybrid quantum-classical
algorithms rely on these circuits (51). The key idea of the variational circuit is optimizing the parameters according
to an objective function. variational quantum circuits consist of two phases (see Figure 1): the quantum phase and
classical phase. The quantum phase includes state preparation, quantum circuit is the heart of the variational circuit
that parameterizes input (X) based on numbers of parameters (θ) and measurement. The classical phase includes the
output of the circuit, objective function and learning algorithm.” quantum variational circuit” can be optimized by
classical optimization algorithms as gradient descent, stochastic gradient descent and particle swarm optimization.
some of the uses of variational quantum circuit such as machine learning (40-41), optimization (52), deep learning
(53) and The variational circuit also uses in solving the complicated optimization problems.

d) Classical proposed quantum classification scheme

We propose a new a quantum classification scheme consists of six phases as collect datasets, pre-processing &
encoding, training, validation (evaluate and optimize model), testing phase and decoding phase. (as shown in Figure
2-part A). Also, we show a classical classification scheme (as shown in Figure 2 -part B), we describe these phases as
following:
 8

Figure 1. Illustration two phases of the variational circuit (parametrized quantum circuit) for Hybrid quantum-classical applications in a quantum
machine learning (QML) and quantum neural networks (QNNS). Quantum phase that includes a quantum circuit that consists of state
preparation, circuit, and measurement. another phase is a classical phase that includes output after measurement and consists of cost function
and learning algorithms which updates parameters (θ).

1. collect datasets
According to the classifier, we can determine whether it works on quantum or classical data. Some of the quantum-
inspired ML works on classical data for instance (37). and other works on quantum data (27). Also, Hybrid quantum-
classical classifiers work on quantum data as (33).

2. Pre-processing & Encoding

The preprocessing phase may be applied by classical or quantum techniques. For example, for handling missing
values, we can use classical or quantum techniques. Then, reduce number of the features to are compatible with
available limited numbers of qubits on quantum device and split data to training, validation and test set. Then, apply
the encoding for mapping data from classical data (X) to quantum data ( |𝜓𝑥 ⟩) in order to prepare data as quantum
states. Next, applying a quantum classifier.

3. Training
Training phase varies according to quantum classifier type. For example, in the case of quantum-inspired classifier,
we train model on classical machine. In the case of quantum kernel classifier, we estimate kernel on a quantum
device) for instance, IBM Q computer or quantum simulator( .

4. Validation
We evaluate and update parameters to decrease cost function and improve performance of model.

5. Testing
We test model performance by confusion matrix and calculate test time.

6. Decoding (or readout)

To calculate output data and translate into classical form (i.e. C = {−1,1} ).

4 | DISCUSSION
Quantum algorithms are quantum models when running a real quantum computer on but don’t run their classical
counterparts on such as Shor’s algorithm (8), Grover’s algorithm (9], addition on a quantum computer (54). for more
details about these algorithms, see (13), (25), (55).
 9

Quantum machine learning (QML) has become a new growth research field and has appeared in many applications
and countries. the progress and success of quantum computing have observed significantly So, we need to apply the
advantages and properties of QM with the machine learning branch.

Figure 2. proposed quantum scheme for classification and prediction pipeline.

Above part (A): the quantum classification and prediction scheme consist of preprocessing, encoding, train, test, and
decoding steps. The quantum classifier requires two main steps encoding and decoding (readout) data. First, the step
of preprocessing & encoding in the quantum model may be applied by classical or quantum techniques. Many
quantum classifiers work on data and others work on classical often in quantum-inspired classifiers. In the case of
quantum inspired classifiers that work on quantum data, we should map classical data (X) to quantum data ( |𝜓𝑥 ⟩)
.thirdly, training phase, this step varies by quantum classifier type. For example, in the case of quantum-inspired, we
train a model on the classical machine. In case of quantum kernel classifier, we kernel on a quantum device. Last,
Decoding (measurement) to convert output labels data to classical. Below part (A): classical classification and
prediction pipeline consists of preprocessing, Training & validation, testing and predication phase.

Due to the success of Shor’s algorithm for solving factorization problems with exponential speedup, Grover’s
algorithm search problem in unstructured data with a quadratic speedup, and HHL algorithm (56) to solve linear
systems equations. Quadratic Unconstrained Binary Optimization problem is NP-hard problems that are hard to the
classical hardware and algorithms solve it. QUBO problem is solved by quantum annealing(57) (see table 3) .

It’s time to discuss the survey questions: Why do we proceed towards Quantum Machine Learning (QML)? To solve
challenges and difficult tasks that still exist in classical machine learning. With increasing size of data, learning phase
training time and memory consumption to increase during the learning phase. to achieve low-cost learning with
higher accuracy and performance is a challenging task with classical algorithms. It’s hard to estimate kernel functions
in higher dimensions (58-60). it hard to determine the eigenvector (60) and solve complex optimization problems.
 10

Problem Solved By Speed up

Factorization Shor’s algorithm (8) Exponential

Searching in unstructured Grover’s algorithm (9) Quadratic

data

Linear systems equations HHL algorithm (56) Exponential

Quadratic Unconstrained D-Wave quantum computer -

Binary Optimization (QUBO) (70)

IBM's Neurosynaptic
hardware (71)

TABLE 3 Problems solved by quantum computing.

According to Dunjko, V et al. (15) Quantum ML can improve in run-time and efficiency. D-wave introduced Quadrant
algorithms based on Central Processing Unit (CPU), Graphics Processing Unit (GPU) and Quantum annealing
computer. Quadrant ML algorithms can achieve more performance with low-cost training data with large data over
classical deep learning and traditional algorithms (61) (62).

Secondly, how does the concept of quantum computing enhance classical machine learning? Quantum computing
can boost traditional algorithms by two different methods: The first method, implement classical algorithms on
quantum computers or simulators. This method needs to encode classical data into quantum data. Another method,
build QML algorithms depend on Quantum algorithms or quantum subroutines (63) such as Amplitude Amplification,
Grover’s algorithm, Quantum matrix-inversion, the quantum phase estimation, Variational quantum (48), quantum
annealing, Sampling (47).

Quantum Fourier Transform (QFT) (64) is the heart of other quantum algorithms such as Quantum Phase Estimation
algorithm and Shor’s algorithm. Quantum Fourier Transform is a quantum version of inverse discrete Fourier
Transform. QFT achieves exponential speed-up. The idea of QFT depends on mapping amplitudes of the current
quantum state (R) to the next amplitudes of the quantum state (Q). 𝑥𝑖
𝑁−1
Where |R⟩= ∑𝑁−1
𝑖=1 𝑥𝑖 |¡ ⟩ , |Q⟩= ∑𝑗=1 𝑦𝑗 |j⟩ with 𝑥𝑖 , 𝑦𝑗 are complex numbers and mapping equation as following
[60,75] :
1 𝑗𝑘
2𝜋𝑖 𝑁
Qj= ∑𝑁−1
𝑛=1 𝑋𝑛 𝑒 (5)
√𝑁

Where j= 1, 2,….,N-1.

Quantum Phase Estimation (QPE) (65, 66) is one of the most important subroutines in many quantum algorithms
[8,63] and quantum machine learning algorithms. QPE algorithm is based on Quantum Fourier Transform (QFT)
subroutine. The goal of Quantum Phase Estimation (QPE) algorithm is to find the eigenvalue ( 𝑒 2𝜋𝑖 𝜃 ) of the unitary
matrix (U) or find the optimal approximation of the phase ( 𝜃) asthe following equation.

U|φ⟩= 𝑒 2𝜋𝑖 𝜃 |φ⟩ (6)

Where 0 ≤ θ < 1 and |φ⟩ is eigenvector.

Amplitude Amplification (AA) (67) (also called Quantum Interference) is one of the fundamental subroutines and is
the key idea for the famous Grover’s algorithm, and Grover’s algorithm is a subroutine for many different algorithms.
The key objective of Amplitude Amplification (AA) is boosting the solution probability of the amplitude (Р) from
 11

arbitrary state to target state overall iterations (68, 69). To make amplify the successful probability of the amplitude
by using the following formula:

2
1- (𝑚3 (Р)) (𝑚2 (Р)) (7)

Where m is the number of iterations and Р is success probability.

From table 4 that summarizes quantum subroutines and applications of each subroutine. We note that quantum
Fourier transform (QFT) and variational quantum circuit is the most common and used subroutines in most
algorithms and applications.

Subroutine Applications Refs.

Amplitude amplification/Grover’s Quantum counting and searching (9,67)

algorithm
sampling Quantum deep learning &” boson (47)
sampling”
Quantum phase estimation Quantum counting (8 ,55)

Quantum Matrix- inversion/HHL QSVM , kernel least squares and (55,56)

Algorithms machine learning
Quantum Fourier Transform Shor’s algorithm, (13,54,64)
Cryptography,
Information processing,
Communication, Discrete logarithm
and
Phase estimation
Variational quantum circuit Optimization, classification (33,48,40,51,53)

Quantum Annealing Eigen state solver (22), (72, 73), (74)

Machine learning
Optimization
Nurse Scheduling
Healthcare

TABLE 4 show quantum subroutines that enhanced classical ML techniques, algorithms applied
based on it. To build QML algorithms and some of subroutines applications.

5 | CHALLENGES AND FUTURE DIRECTIONS

In this section, we outline and present challenges and future directions in QML as small-scale quantum computer,
Limited quantum bits, encoding methods and develop a new QML techniques.

1. Small-scale quantum computer

the big and key challenge is to build quantum computers with a large number of qubits to implement, test QML
algorithms and work with large different data in the near future. As shown in Figure 3, numbers of qubits achieved
by different technology companies (75) such as Rigetti, IBM, Q-Wave, Xanadu, Google, and Microsoft. Up to now, the
quantum computer is developed in small-scale and this restricts us use a limited amount of data (76). So, researchers
 12

develop algorithms work on available small-scale and “noisy intermediate-scale quantum“ (NISQ) quantum hardware
compatible with number of qubits hardware(77).limited qubits number using a less number of features that lead to
losing a lot of important data and with limited qubits can’t apply big data processing on quantum devices (78, 79) .

Progress of quantum computer & Numbers of

140
Quantum Bits 128
120

100

80
72
Qubits

60
50 49
40

20

0 2 3 5
IBM, THE RIGETTI IBM 2016 IBM 2017 INTEL 2018 GOOGLE RIGETTI
MIT MEDIA 2016 2018 2019
LAB, AND
UC
BERKELEY Quantum Computer
1997

Figure 3. A chart shows the progress of quantum computer with numbers Quantum Bits: Source (75), Numbers of
qubits achieved by different companies. The first quantum computer with 2-qubits build by IBM, the MIT Media Lab,
and UC Berkeley in 1997.IBM achieved in 2017, 50 qubits.20 years to increase 48 qubits .in 2019, Rigetti achieved
128 qubits.in 2 years 78 qubits increased. If numbers of qubits increase with a high error rate don’t make a powerful
quantum processor (80). Powerful quantum processer with a low error rate will increase exponentially by increasing
numbers of quantum bits.

2. Encoding methods
Encoding data to quantum states is one of QML challenges, this process takes high consumption time and power for
mapping from classical data into quantum data (74) (i.e. image and big data). So, design new techniques for encoding
data is an interesting future direction for research. In 2020, LaRose et al (81).presented a binary quantum classifier
in order to robust with noise quantum states based on the selection of the best encoding method to load data into
the quantum system. Furthermore, the authors discussed different methods for encoding with quantum binary
classifier. also, the authors applied several encoding data on the same data and showed that encoding data
techniques can be improved the model accuracy. In 2020, the authors (74) introduced a new method for encoding
large amount of image data into a few number of qubits after compression to reduce of image based on “quantum
annealing computer. As well, the authors trained “restricted boltzmann machine” to classify images data.

3. A new QML techniques

In near future, we think that researchers will develop new theoretical and applied ML algorithms that are compatible
with available quantum hardware. using quantum information theory or quantum subroutines for solving ML
problems and improve performance. As well, create versions for current algorithms in several fields. For example,
quantum neural networks (QNNs), quantum deep Learning, quantum-inspired ML, and quantum-enhanced ML(82).in
addition to, develop new classification techniques based on quantum “variational circuit” and working on reducing
the depth of quantum circuit. And also, implementation of classical techniques on real quantum machines or
 13

simulators. using quantum annealing and adiabatic computation with classical ML to develop new QML paradigms
and solve complicated optimization problems related to ML problems.

6 | APPLICATIONS

Quantum Machine Learning (QML) techniques are more effective in many real-world applications comparable to
traditional machine learning in speedup and accuracy as big data classification [24], forecasting series, spam
detection, image compression, medical domain (30,43,37) (i.e., cervical cancer detection (83), electronic calculations
(84) ,decision games (85), natural language processing (NLP), recommendation systems (86, 87) , speech recognition
(88), image classification (27,44) , and electrocardiogram signals classification (89) .Applications related to hybrid
quantum-classical approach such as scheduling problems (90), and classification task (40-41).

7 | CONCLUSION

This survey paper reviewed the most recent articles for quantum machine learning techniques for various problems
that improve and handle different problems more effectively and accurately from traditional techniques in
conventional computing. And also, we compared between quantum-inspired and hybrid- quantum-classical
algorithms. as mentioned in section 2. (literature review), QML algorithms outperformed classical ML in performance
and speed up. In addition, we presented the quantum scheme for classification problems or in general in supervised
learning then we outline two methods for mapping data. We discussed methods of quantum to enhance machine
learning such as quantum subroutines. We presented applications of QML in different fields and future vision, to
open new avenues for use of quantum machine learning in the research field. Our future work, we will apply
quantum-inspired and hybrid quantum-classical algorithm on real-world data to compare its performance.

ACKNOWLEDGEMENTS
We are grateful to prof. Giuseppe Sergioli for helpful discussion.

REFERENCES

1. Alpaydin E. Introduction to machine learning: MIT press; 2020.

2. Mehta P, Bukov M, Wang C-H, Day AG, Richardson C, Fisher CK, et al. A high-bias, low-variance introduction to
machine learning for physicists. Physics reports. 2019.
3. Kubat M. An introduction to machine learning: Springer; 2017.
4. Biamonte J, Wittek P, Pancotti N, Rebentrost P, Wiebe N, Lloyd S. Quantum machine learning. Nature.
2017;549(7671):195-202.
5. Broadbent A, Schaffner C. Quantum cryptography beyond quantum key distribution. Designs, Codes and
Cryptography. 2016;78(1):351-82.
6. Dunjko V, Briegel HJ. Machine learning & artificial intelligence in the quantum domain: a review of recent progress.
Reports on Progress in Physics. 2018;81(7):074001.
7. Farouk A, Tarawneh O, Elhoseny M, Batle J, Naseri M, Hassanien AE, et al. Quantum computing and cryptography:
an overview. Quantum Computing: An Environment for Intelligent Large Scale Real Application: Springer; 2018. p.
63-100.
8. Shor PW. Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer.
SIAM review. 1999;41(2):303-32.
9. Grover LK, editor A fast quantum mechanical algorithm for database search. Proceedings of the twenty-eighth
annual ACM symposium on Theory of computing; 1996.
10. Schuld M, Sinayskiy I, Petruccione F. An introduction to quantum machine learning. Contemporary Physics.
2015;56(2):172-85.
11. Wittek P. Quantum machine learning: what quantum computing means to data mining: Academic Press; 2014.
 14

12. Sharma S. QEML (Quantum Enhanced Machine Learning): Using Quantum Computing to Enhance ML Classifiers
and Feature Spaces. arXiv preprint arXiv:200210453. 2020.
13. Coles PJ, Eidenbenz S, Pakin S, Adedoyin A, Ambrosiano J, Anisimov P, et al. Quantum algorithm implementations
for beginners. arXiv preprint arXiv:180403719. 2018.
14. Nakahara M, Wan Y, Sasaki Y. Diversities in Quantum Computation and Quantum Information: World Scientific;
2012.
15. Dunjko V, Taylor JM, Briegel HJ. Quantum-enhanced machine learning. Physical review letters.
2016;117(13):130501.
16. Aïmeur E, Brassard G, Gambs S, editors. Machine learning in a quantum world. Conference of the Canadian
Society for Computational Studies of Intelligence; 2006: Springer.
17. Montanaro A. Quantum algorithms: an overview. npj Quantum Information. 2016;2(1):1-8.
18. Jeswal S, Chakraverty S. Recent developments and applications in quantum neural network: A review. Archives
of Computational Methods in Engineering. 2019;26(4):793-807.
19. Benedetti M, Lloyd E, Sack S, Fiorentini M. Parameterized quantum circuits as machine learning models.
Quantum Science and Technology. 2019;4(4):043001.
20. Resch S, Karpuzcu UR. Quantum computing: an overview across the system stack. arXiv preprint
arXiv:190507240. 2019.
21. Ciliberto C, Herbster M, Ialongo AD, Pontil M, Rocchetto A, Severini S, et al. Quantum machine learning: a classical
perspective. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences.
2018;474(2209):20170551.
22. Willsch D, Willsch M, De Raedt H, Michielsen K. Support vector machines on the D-Wave quantum annealer.
Computer Physics Communications. 2020;248:107006.
23. Headquarters C. Technical Description of the D-Wave Quantum Processing Unit. 2019.
24. Rebentrost P, Mohseni M, Lloyd S. Quantum support vector machine for big data classification. Physical review
letters. 2014;113(13):130503.
25. da Silva AJ, Ludermir TB, de Oliveira WR. Quantum perceptron over a field and neural network architecture
selection in a quantum computer. Neural Networks. 2016;76:55-64.
26. Schuld M, Sinayskiy I, Petruccione F. Prediction by linear regression on a quantum computer. Physical Review A.
2016;94(2):022342.
27. Tiwari P, Melucci M. Towards a quantum-inspired binary classifier. IEEE Access. 2019;7:42354-72.
28. Sergioli G, Giuntini R, Freytes H. A new quantum approach to binary classification. PloS one. 2019;14(5).
29. Ding C, Bao T-Y, Huang H-L. Quantum-Inspired Support Vector Machine. arXiv preprint arXiv:190608902. 2019.
30. Sergioli G, Russo G, Santucci E, Stefano A, Torrisi SE, Palmucci S, et al. Quantum-inspired minimum distance
classification in a biomedical context. International Journal of Quantum Information. 2018;16(08):1840011.
31. Dang Y, Jiang N, Hu H, Ji Z, Zhang W. Image classification based on quantum K-Nearest-Neighbor algorithm.
Quantum Information Processing. 2018;17(9):239.
32. Chen H, Gao Y, Zhang J. Quantum k-nearest neighbor algorithm. Dongnan Daxue Xuebao. 2015;45(4):647-51.
33. Lu S, Braunstein SL. Quantum decision tree classifier. Quantum information processing. 2014;13(3):757-70.
34. Bishwas AK, Mani A, Palade V. Big Data Quantum Support Vector Clustering. arXiv preprint arXiv:180410905.
2018.
35. Casaña-Eslava RV, Lisboa PJ, Ortega-Martorell S, Jarman IH, Martín-Guerrero JD. A Probabilistic framework for
Quantum Clustering. arXiv preprint arXiv:190205578. 2019.
36. Yu C-H, Gao F, Wen Q. An improved quantum algorithm for ridge regression. IEEE Transactions on Knowledge
and Data Engineering. 2019.
37. Sagheer A, Zidan M, Abdelsamea MM. A novel autonomous perceptron model for pattern classification
applications. Entropy. 2019;21(8):763.
38. Adhikary S, Dangwal S, Bhowmik D. Supervised learning with a quantum classifier using multi-level systems.
Quantum Information Processing. 2020;19(3):89.
39. Chakraborty S, Shaikh SH, Chakrabarti A, Ghosh R. A hybrid quantum feature selection algorithm using a quantum
inspired graph theoretic approach. Applied Intelligence. 2020:1-19.
 15

40. Havlíček V, Córcoles AD, Temme K, Harrow AW, Kandala A, Chow JM, et al. Supervised learning with quantum-
enhanced feature spaces. Nature. 2019;567(7747):209-12.
41. Schuld M, Killoran N. Quantum machine learning in feature Hilbert spaces. Physical review letters.
2019;122(4):040504.
42. Mitarai K, Negoro M, Kitagawa M, Fujii K. Quantum circuit learning. Physical Review A. 2018;98(3):032309.
43. Pointing J. Quantum Algorithm for Handling Missing Data. harvard. 2018.
44. Ruan Y, Xue X, Liu H, Tan J, Li X. Quantum algorithm for k-nearest neighbors classification based on the metric of
hamming distance. International Journal of Theoretical Physics. 2017;56(11):3496-507.
45. Grant E, Benedetti M, Cao S, Hallam A, Lockhart J, Stojevic V, et al. Hierarchical quantum classifiers. npj Quantum
Information. 2018;4(1):1-8.
46. Zhang D-B, Zhu S-L, Wang Z. Nonlinear regression based on a hybrid quantum computer. arXiv preprint
arXiv:180809607. 2018.
47. Benedetti M, Garcia-Pintos D, Perdomo O, Leyton-Ortega V, Nam Y, Perdomo-Ortiz A. A generative modeling
approach for benchmarking and training shallow quantum circuits. npj Quantum Information. 2019;5(1):1-9.
48. Schuld M, Bocharov A, Svore KM, Wiebe N. Circuit-centric quantum classifiers. Physical Review A.
2020;101(3):032308.
49. Gambs S. Quantum classification. arXiv preprint arXiv:08090444. 2008.
50. Schuld M, Petruccione F. Supervised learning with quantum computers: Springer; 2018.
51. McClean JR, Romero J, Babbush R, Aspuru-Guzik A. The theory of variational hybrid quantum-classical algorithms.
New Journal of Physics. 2016;18(2):023023.
52. Moll N, Barkoutsos P, Bishop LS, Chow JM, Cross A, Egger DJ, et al. Quantum optimization using variational
algorithms on near-term quantum devices. Quantum Science and Technology. 2018;3(3):030503.
53. Chen SY-C, Goan H-S. Variational Quantum Circuits and Deep Reinforcement Learning. arXiv preprint
arXiv:190700397. 2019.
54. Draper TG. Addition on a quantum computer. arXiv preprint quant-ph/0008033. 2000.
55. Hales L, Hallgren S, editors. An improved quantum Fourier transform algorithm and applications. Proceedings
41st Annual Symposium on Foundations of Computer Science; 2000: IEEE.
56. Harrow AW, Hassidim A, Lloyd S. Quantum algorithm for linear systems of equations. Physical review letters.
2009;103(15):150502.
57. Lewis M, Glover F. Quadratic unconstrained binary optimization problem preprocessing: Theory and empirical
analysis. Networks. 2017;70(2):79-97.
58. Silverman B. Density estimation for statistics and data analysis.[EPub]. New York: Routledge; 2018.
59. Hansen BE. Lecture notes on nonparametrics. Lecture notes. 2009.
60. Wang W, Yang N, Zhang Y, Wang F, Cao T, Eklund P. A review of road extraction from remote sensing images.
Journal of traffic and transportation engineering (english edition). 2016;3(3):271-82.
61. Li RY, Di Felice R, Rohs R, Lidar DA. Quantum annealing versus classical machine learning applied to a simplified
computational biology problem. NPJ quantum information. 2018;4(1):1-10.
62. Systems D-W. D-Wave Announces Quadrant Machine Learning Business Unit. 2018.
https://www.dwavesys.com/press-releases/d-wave-announces-quadrant-machine-learning-business-unit
63. Schuld M, Petruccione F. Quantum Machine Learning. 2017.
64. Coppersmith D. An approximate Fourier transform useful in quantum factoring. arXiv preprint quant-
ph/0201067. 2002.
65. Luis A, Peřina J. Optimum phase-shift estimation and the quantum description of the phase difference. Physical
review A. 1996;54(5):4564.
66. Cleve R, Ekert A, Macchiavello C, Mosca M. Quantum algorithms revisited. Proceedings of the Royal Society of
London Series A: Mathematical, Physical and Engineering Sciences. 1998;454(1969):339-54.
67. Brassard G, Hoyer P, Mosca M, Tapp A. Quantum amplitude amplification and estimation. Contemporary
Mathematics. 2002;305:53-74.
68. Brassard G, Høyer P, Tapp A, editors. Quantum counting. International Colloquium on Automata, Languages, and
Programming; 1998: Springer.
 16

69. Ambainis A, editor Variable time amplitude amplification and quantum algorithms for linear algebra
problems2012.
70. Neukart F, Compostella G, Seidel C, Von Dollen D, Yarkoni S, Parney B. Traffic flow optimization using a quantum
annealer. Frontiers in ICT. 2017;4:29.
71. Alom MZ, Van Essen B, Moody AT, Widemann DP, Taha TM, editors. Quadratic Unconstrained Binary
Optimization (QUBO) on neuromorphic computing system. 2017 International Joint Conference on Neural Networks
(IJCNN); 2017: IEEE.
72. Mott A, Job J, Vlimant J-R, Lidar D, Spiropulu M. Solving a Higgs optimization problem with quantum annealing
for machine learning. Nature. 2017;550(7676):375-9.
73. Adachi SH, Henderson MP. Application of quantum annealing to training of deep neural networks. arXiv preprint
arXiv:151006356. 2015.
74. Sahner D. A Potential Role for Quantum Annealing in the Enhancement of Patient
Outcomes?2018.
https://www.dwavesys.com/sites/default/files/Sahner.2018.pdf
75. Marcos Allende López , Silva MD. Quantum Technologies : Digital transformation, social impact, and cross-sector
disruption2019. 14-5 p.
https://publications.iadb.org/publications/english/document/Quantum_Technologies_Digital_Transformation_Soci
al_Impact_and_Crosssector_Disruption.pdf
76. Herbster M, Mountney P, Piat S, Severini S. Data encoding and classification. Google Patents; 2020.
77. Preskill J. Quantum Computing in the NISQ era and beyond. Quantum. 2018;2:79.
78. Perdomo-Ortiz A, Benedetti M, Realpe-Gómez J, Biswas R. Opportunities and challenges for quantum-assisted
machine learning in near-term quantum computers. Quantum Science and Technology. 2018;3(3):030502.
79. Shiba K, Sakamoto K, Yamaguchi K, Malla DB, Sogabe T. Convolution filter embedded quantum gate autoencoder.
arXiv preprint arXiv:190601196. 2019.
80. Gambetta JM, Chow JM, Steffen M. Building logical qubits in a superconducting quantum computing system. npj
Quantum Information. 2017;3(1):1-7.
81. LaRose R, Coyle B. Robust data encodings for quantum classifiers. arXiv preprint arXiv:200301695. 2020.
82. Dunjko V, Wittek P. A non-review of Quantum Machine Learning: trends and explorations. Quantum Views.
2020;4:32.
83. Iliyasu AM, Fatichah C. A quantum hybrid PSO combined with fuzzy k-NN approach to feature selection and cell
classification in cervical cancer detection. Sensors. 2017;17(12):2935.
84. Xia R, Kais S. Quantum machine learning for electronic structure calculations. Nature communications.
2018;9(1):1-6.
85. Clausen J, Briegel HJ. Quantum machine learning with glow for episodic tasks and decision games. Physical
Review A. 2018;97(2):022303.
86. Tang E, editor A quantum-inspired classical algorithm for recommendation systems. Proceedings of the 51st
Annual ACM SIGACT Symposium on Theory of Computing; 2019.
87. Kerenidis I, Prakash A. Quantum recommendation systems. arXiv preprint arXiv:160308675. 2016.
88. Kafian S, Yaghoobi M, Attari I. P65: Speech Recognition Based on Bbrain Signals by the Quantum Support Vector
Machine for Inflammatory Patient ALS. The Neuroscience Journal of Shefaye Khatam. 2018;6(2):96-.
89. Tang X, Shu L. Classification of electrocardiogram signals with RS and quantum neural networks. International
Journal of Multimedia and Ubiquitous Engineering. 2014;9(2):363-72.
90. Tran TT, Do M, Rieffel EG, Frank J, Wang Z, O'Gorman B, et al., editors. A hybrid quantum-classical approach to
solving scheduling problems. Ninth annual symposium on combinatorial search; 2016.