Qubit-Based Framework for Quantum Machine Learning:

Bridging Classical Data and Quantum Algorithms

Bhavna Bose and Saurav Verma

Mukesh Patel School of Technology Management and Engineering, SVKM’s NMIMS

This paper dives into the exciting and ers a clear and comprehensive introduction
rapidly growing field of quantum com- to quantum computing, its role in machine
puting, explaining its core ideas, current learning, and the immense possibilities it
progress, and how it could revolutionize holds for the future of technology.
the way we solve complex problems. It
arXiv:2502.11951v1 [cs.CE] 17 Feb 2025

starts by breaking down the basics, like

qubits, quantum circuits, and how princi- 1 Introduction
ples like superposition and entanglement
make quantum computers fundamentally Machine learning has transformed the data an-
different—and far more powerful for cer- alytics industry like never before. With avail-
tain tasks—than the classical computers ability of a variety of data in large volumes at
we use today. We also explore how quan- high velocity, i.e., ’Big Data’, significant insights
tum computing deals with complex prob- can be drawn. As the volume of data increases,
lems and why it’s uniquely suited for chal- so does the difficulty in processing it. There is
lenges classical systems struggle to handle. a need for systems that can efficiently process
A big part of this paper focuses on Quan- large data sets, using complex algorithms, in real
tum Machine Learning (QML), where the time and with high accuracy. In our fast-paced
strengths of quantum computing meet the world, the industry is driven by the urge to speed
world of artificial intelligence. By process- up processed and make them more efficient and
ing massive datasets and optimizing in- safe. Quantum computing provides a platform
tricate algorithms, quantum systems of- to achieve this, with a few caveats! Quantum
fer new possibilities for machine learn- physics has been around for quite some time.
ing. We highlight different approaches to However, Shor’s algorithm [49] catapulted the in-
combining quantum and classical comput- terest in Quantum Computing. Fig. 1 shows the
ing, showing how they can work together progress of Quantum research from 1982 to 2024.
to produce faster and more accurate re- Quantum computing research has grown in
sults. Additionally, we explore the tools leaps and bounds.Google and IBM are the two
and platforms available—like TensorFlow front runners in this Quantum race. In just a
Quantum, Qiskit, and PennyLane—that couple of years IBM has successfully developed
are helping researchers and developers a quantum computer capable of running up to
bring these theories to life. Of course, 5000 two qubit gate operations. [32] . In Decem-
quantum computing isn’t without its hur- ber 2024, Google launched their breakthrough
dles. Challenges like scaling up hardware, 105 Qubit Quantum chip called Willow.[1]. This
correcting errors, and keeping qubits sta- latest development has tackled a very important
ble are significant roadblocks. Yet, with challenge in Quantum computing of error reduc-
rapid advancements in cloud-based plat- tion and decoherence as we increase the number
forms and innovative technologies, the po- of Qubits. These advancements have accelerated
tential of quantum computing feels closer the growth in Quantum research and develop-
than ever. This paper aims to offer read- ment.
Consider the performance of an algorithm as a
Bhavna Bose: [email protected] function of the amount of time spent in solving
Saurav Verma: [email protected]
the problem. As size of the input increases, high

1
 1900 1935 1964 1970 1981 1982
Plank’s EPR Bell’s Birth of Feynman’s Fractional
Quantum Paradox Inequality Quantum Quantum Quantum
Hypothesis Information Computer Hall (FQH)
Theory Proposal state was Quantum
discovered Cryptography
1984 (BB84 protocol)
by IBM

2004 1996 1994 1994 1985

Circuit DiVincenzo’s Grover’s Shor’s Deutsch’s
QED Demo Criteria for Algorithm Algorithm Universal
Quantum Quantum
Computers Computer
The
transmon
2007 supercon-
ducting
qubit
2007 2013 2016 2019 2020 2024 2024
D Wave one Rigetti Microsoft Google IBM 5000 two Quantum
quantum Computing Station Q Quantum Quantum Qubit chip, Willow
annealer Supremacy Roadmap processor by by Google
IBM

Figure 1: Quantum computing timeline

degree polynomial functions grow much slower the Dirac notation ketψ written as |ψ⟩. Each sin-
than quadratic functions as shown in Fig 2. gle qubit stores information about two states 0
Quantum systems provide speedups in this area. and 1. However, on measurement, the qubit col-
lapses to a single state i.e. 0 or 1. This means
1037 that in the intermediate process fewer qubits (N)
N
can be used to process more (2N )data. All this is
N log(N )
possible due to the duality of a quantum particle.
1028 N3
Time/Memory

N6 This does not mean that classical computers

2N will become obsolete. It is quite the opposite!
1019 The current research (as we will see the next sec-
6N
tion) is working on a symbiotic interaction be-
tween Quantum computers and Classical comput-
1010 ers. Also as seen from Fig. 2, quantum comput-
ing offers speedups for a particular class of prob-
lems. Each processing paradigm has its strength
101 and by harnessing the strength of both systems,
10 20 30 40 50 solutions of traditionally unsolvable problem can
# Elements (Input Size) be attempted.
Comparison of time complexities based on Just as quantum computing offers unique so-
Figure 2:
number of inputs lutions to many unsolvable problems, quantum
computing faces its unique challenges. [51], dis-
Classical systems store single information in a cusses the many challenges faced in the devel-
single bit i.e. 0 or 1. In contrast, quantum com- opment of quantum computing based systems.
puters store information in qubits shown using These include hardware scalability, decoherence,

2
and limited gate operations to name a few. Even QML and suggesting resources for further explo-
with the tremendous work done in the develop- ration. Rather than striving for exhaustive cov-
ment of quantum hardware by industry giants like erage, the goal is to equip readers with sufficient
IBM and Google, we are at least half a decade background and guidance for deeper study.
away before practical quantum systems will be This paper is organized as follows: Section 2 in-
available. troduces the basic concepts of quantum comput-
ing starting from Qubits till problem complex-
1.1 Related Literature ity. Section 3 discusses quantum computation
with an introduction to Quantum Circuits and
The work done in [10, 36, 48] gives an idea about their modeling. Section 4 is briefly discusses the
different applications of machine learning, quan- steps involved in machine leaning, while Section
tum information processing and the overlap be- 5 dives into quantum machine learning discussing
tween these two fields. For those seeking an entry the phases/steps to collaboratively use classical
point into the topics of Quantum Machine Learn- and quantum techniques. Section 6 discusses an
ing (QML), [23] serves as a valuable resource, implementation framework for of Quantum ma-
while [15] emphasizes algorithmic complexity and chine learning algorithms. Section 7 gives an
theoretical foundations through illustrative ex- overview of some quantum computing platforms
amples of QML techniques. available currently. Finally, Section 8 concludes
Further exploration into Quantum Neural Net- the paper.
works (QNN) is provided by [14] and [35], which
compare key QNN architectures, summarize their
characteristics, and offer broader context within 2 Introduction to Quantum Comput-
this domain. Additionally, [22] demonstrates ex- ing
amples of quantum enhancements to Machine
Learning and Artificial Intelligence. Among in- In this section we will discuss the building block
troductory texts, the monograph by [56] stands of Quantum systems, Qubits and the problem
out as the most comprehensive resource, pre- types which can be solved using the Quantum
senting significant discoveries in Quantum Ma- approach.
chine Learning alongside a thorough introduc-
tion to both classical machine learning and quan- 2.1 Qubits
tum computing. Similarly, [47] delves into
Qubits, are the basic building block of a quan-
well-established techniques, offering a detailed
tum computing system. Qubits in quantum com-
overview of the field.
puters are analogous to bits in classical systems.
A unique perspective is provided by [54], which
Qubits are quantum particles and can be imple-
explores QML’s applications in robotics while
mented using many physical particles like super-
also addressing classical QML concepts. Recent
conducting qubits, trapped ion qubits, quantum
works have begun to propose specific application
dots, photons, neutral atoms to name a few.
scenarios for QML. For instance, [46] discusses
The different types of qubits used in quantum
how novel QNN architectures could address nu-
computing today are listed below:
merical challenges in finance, while [58] exam-
ines the potential of quantum neural networks • Superconducting qubits: Superconducting
to enhance medical image recognition. Physics materials like aluminum, niobium and tan-
remains the most prominent area for QML ap- talum are used to make the qubits. As ex-
plications, with [13] offering a comprehensive re- tremely low temperatures, they enable fast
view. Conversely, other cross-disciplinary appli- and fine tuned computations.
cations have seen less significant contributions
• Trapped ion qubits: Qubits implemented as
from QML, as noted by [45].
trapped ion particles can stay for longer peri-
This work aims to an introduction to Quantum
ods without being disrupted by environmen-
Machine Learning (QML). It covers foundational
tal noise and less errors in measurements.
concepts, classical Quantum approaches integrat-
ing the existing systems with Quantum mecha- • Quantum dots: they are nanoscale semicon-
nisms, a proposed framework for implementing ductor particles. The electron in a Quantum

3
 Dot and confined in a very small three di- the time and space required to compute a prob-
mensional space making their energy levels lem which is normally a function of the number
discrete and measurable. of inputs i.e. input size. Fig. 3 shows the set of
decision problems that require a certain amount
• Photons: Photons are light particles which of time and space.
exhibit the phenomena of entanglement and
superposition. Also, due to their weak inter- • Exponential Space Problems (EXPSPACE) :
actions with surrounding environments they The space required to solve problems in this
more immune to noise. class is an exponential fuction of the number
of inputs.
• Neutral atoms: Neutral atoms are atoms
• Exponential Time Problems (EXPTIME) :
with equal number of electrons and protons,
The time required to solve problems in this
making them stabe for quantum communi-
class is an exponential fuction of the number
cations. Quantum information is encoded in
of inputs.
the internal states of these atoms.
• Polynomial Space Problems (PSPACE) :
Table 1 gives a summary of the different types The space required to solve problems in this
of qubits and the companies building Quantum class is a polynomial function of the number
Processing Units (QPUs) using that technology. of inputs.
The spin and amplitude of the qubits carry in-
formation. To achieve this, the qubits have to be • Polynomial time problems (P ): Problems in
kept at very low temperatures, just above abso- this class can be solved in polynomial time
lute zero (-459 degrees Fahrenheit), to be able to as the input size increases.
control their behavior (i.e. spin and amplitude). • Non deterministic Polynomial time Prob-
The main challenge to quantum computing is the lems (NP): Problems in the category cannot
difficulty in maintaining such low temperatures. be solved in polynomial time. However, a
Even with a slight increase in temperature, the solution to these problems can be verified in
quantum particle (qubit) with change its state polynomial time.
i.e. loose its value also called decoherence.[34]
Classical bits do not loose their information i.e • NP-complete problems: Problems in this
they maintain their state over time. However, class are the most difficult problems in NP
qubits loose their state very quicky i.e. loose co- and have no known polynomial solution.
herence in less that 300µs ![7] The low coherence This is where famous problems like the trav-
implies that all computations must be completed eling salesman and the game Soduku live.
before the qubit becomes decoherent. Decoher-
ence forces the developers to create circuits and • Bounded-error Polynomial Problems (BPP):
algorithms which will come to the final result very which can be solved within some error
fast and accurately which aligns to the need of the threshold by a probabilistic classical com-
the consumers too. It is a win-win situation ! puter in polynomial time.

• Bounded-error Quantum Polynomial Prob-

2.2 Classical Vs Quantum Complexity lems (BQP): This is the quantum equivalent
of BPP. It is the class of decision problems
As discussed in the previous section, problems solvable by a quantum computer in polyno-
which can be solved efficiently with the classi- mial time with a small chance of error.
cal computers, do not require to migrate to the
more computationally expensive quantum sys-
tems. Then the question arises , which kind of 3 Quantum Logic
problems are apt for quantum computations ? To
answer the question, computational complexity of Quantum computing paradigm has certain core
the problems needs to be explored. This section differences which makes it attractive. Due to
will discuss the different computation complex- the concepts of entanglement, superposition and
ity classes. Computational complexity refers to superdense coding, quantum algorithms provide

4
 Table 1: Qubit technology and Companies working on building QPUs using that technology.

IBM built the Eagle quantum computer with 127 qubits. [33]
Superconducting qubits Google’s Sycamore [2]
Rigetti’s Aspen-M-2 [19]
QuantWare’s built Crescendo and Soprano [41]
Quantinuum [40]
Trapped ion qubits IonQ [29]
Alpine Quantum Technologies [55]
Oxford Ionics [28]
Diraq [21]
Quantum dots Quobly [42]
Quantum Motion [37]
Xanadu [57]
Photons ORCA Computing [17]
QuantumComputing Inc [26]
PsiQuantum [39]
Pasqal [38]
Neutral atoms Atom Computing [16]
QColdQuanta [27]
QuEra [18]

significant speedups to solve problems considered 3.1 Quantum Circuits

unsolvable earlier. As opposed to bits, qubits can
Classical computations are performed by apply-
be evaluated to a 0 or a 1 based on the probabil-
ing logic gates to the input bits. The logic gates
ity of geting the two values, i.e. a qubit is a linear
transform the state of the bits based on certain
function of 0 and 1 values. A qubit is represented
rules.
by the equation Eq. 1
Analogously, gates are applied to qubits to per-
form certain computations and give results.
Much like classical logic gates, a quantum logic
|ψ⟩ = α |0⟩ + β |1⟩ (1) gate transforms, the input A to A′ based on a
transformation function applied to it through the
logic gate. We usually denote a quantum state
where α, β ∈ C are the popabilites of eval- as |ψ⟩ = α |0⟩ + β |1⟩. When the quantum logic
uating a qubit to 0 or 1 repectively. Thus gate is applied to the quantum input state |ψ⟩, it
|α|2 + |β|2 = 1. This means that during the transforms to |ψ ′ ⟩ = α |0′ ⟩ + β |1′ ⟩ . As this is a
intermediate computations, a qubit can be in a valid quantum state, the condition |α′ |2 + |β ′ |2 =
combination of |0⟩ or |1⟩ . Once measured, the 1 is satisfied. [56].
Qubit will evaluate to either a 0 or 1 classical bit Listed in Table. 2. are few quantum gates with
value. their symbols and transformation matrices.

In an n-qubit system, there are 2n probabil-

ity amplitudes, which might suggest an enormous 3.2 Quantum Circuit Model
amount of information. The reality is that the The key components of a quantum circuit as il-
measurements we can perform limit how much lustrated in Fig. 4 are:
information we can actually extract. This idea is
captured by Holevo’s bound, which tells us that n 1. Classical Resources: Quantum circuit can
qubits can encode, at most, n bits of information. be considered to be divided into two parts:
[25]. Classical part and Quantum part. The true

5
 Exponential Space problems [EXPSPACE]

Exponential Time problems [EXPTIME]

Polynomial Space problems [PSPACE]

NP Hard

NP Complete

NP

BQP P

Figure 3: Complexity Classes

strength of a quantum system is harnessed basis states |x1 , x2 , ..., xn ⟩ are prepared in at-
by leveraging the strengths of both the com- most n steps.
putation paradigms and integrating both the
systems.
4. Apply gates and compute: Once the sys-
2. State Space: Similar to bits in classical sys- tem is in its initial state, Quantum gates are
tems, quantum computers work on a number applides to the required Qubits as per the
of qubits. The qubits occupy a state space model/ problem solution and computations
, which is a 2n dimensional Hilbert Space, are preformed.
where n is the number of Qubits.
A computational basis state |x⟩ corresponds 5. Measure the results: After the computa-
to the binary number x where tions, the results are measured in the com-
putational basis state |xi ⟩. The results are
|x⟩ = |x1 , x2 , ..., xn ⟩ , measured classically as xi corresponding to
x = x1 x2 ..nxn and |xi ⟩
xi = 0, 1

3. Prepare the state space: The computational Model is illustrated in Fig.4

6
 Table 2: Quantum Gates, Symbols, Matrices and Resultant Equation
Input Qubit is |ψ⟩ = α |0⟩ + β |1⟩

Gate Name Symbol Matrix Representation Equation

" #
|ψ⟩ H |ψ ′ ⟩ 1 1 α+β α−β
Hadamard √1
2 1 −1
|ψ ′ ⟩ = √
2
|0⟩ + √
2
|1⟩
" #
|ψ⟩ X |ψ ′ ⟩ 0 1
Pauli-X |ψ ′ ⟩ = α |1⟩ + β |0⟩
1 0
" #
|ψ⟩ Y |ψ ′ ⟩ 0 −i
Pauli-Y |ψ ′ ⟩ = iα |1⟩ − iβ |0⟩
i 0
" #
|ψ⟩ Z |ψ ′ ⟩ 1 0
Pauli-Z |ψ ′ ⟩ = α |0⟩ − β |1⟩
0 −1

Preperation of state space

Binary represen- Prepare the computational ba-
tation of n bits sis state |x⟩ = |x1 , x2 ...xn ⟩
x = x1 x2 ...xn in the 2n dimensional
complex Hilbert space

Apply Quantum gates :

Based on the porblem at
hand, apply the combination
of Quantum gates to |x⟩

Measure the results

Measure the results classical Get Reuslts:
i.e convert |y⟩ = y1 , ..., yn The input state |x⟩ gets trans-
to the corresponding formed to the output stats |y⟩
binary y = y1 ...yn

Figure 4: Quantum Circuit Model

4 Machine Learning ing the probability of getting to a particular state.

Quantum computers due their strength in han-
Machine learning tries to mimic the human na- dling vast quantities of data and parallel process-
ture of learning from past experiences to react to ing [31] by Deutch Algorithm, can aid in machine
new stimuli. What this means is, based all the learning tasks. The power of Quantum comput-
past data collected, a model is developed which ing can be leveraged in Machine learning to give
then gives a prediction of the outcome. The 4 combinations of approaches: The work done in
prediction is probabilistic and rarely 100% accu- [30] discusses these 4 approaches as shown in Ta-
rate. Machine learning involves seven steps[31] as ble 3:
shown in Fig. 5 .
The choice of using the classical approach
or quantum approach requires some analysis.
5 Quantum Machine Learning(QML) As mentioned before, quantum computing al-
gorithms give a clear advantage over classical
5.1 Introduction to QML systems for algorithms with very high sample
complexity[47] . Quantum approaches due to
Machine learning relies heavily on probabilities. their probabilistic nature sometimes are more ro-
This is what makes Quantum systems a good can- bust to noise than classical algorithms. [12, 20].
didate for machine learning application as Quan- Some metrics used in quantum - classical model
tum Computing also, basically relies on calculat- selection are computational complexity, sample

7
 Table 3: Different approaches to QML

Approach Concept Working Example

They run on normal

These are regular computers and work
Classical- algorithms that with everyday data, Algorithms de-
Classical take inspiration but use ideas from signed based
(CC): from quantum quantum computing on quantum
mechanics. to improve concepts, like
performance. optimization
techniques that
mimic quantum
behavior but
run on classical
systems.

Quantum
Here, quantum algorithms, or
Classical- computing is adaptations of A quantum ver-
Quantum used to process existing machine sion of a neural
(CQ): traditional learning algorithms, network that pro-
(classical) data. are applied to cesses data like
classical data in a images or text,
way that aims to hoping to solve
outperform regular problems faster
computer than classical
algorithms. computers.

Quantum computers
generate complex
Classical machine Using traditional
Quantum- quantum data, and
learning is used machine learning
Classical then classical
to make sense of techniques to pre-
(QC): machine learning
quantum data. dict the behav-
algorithms are used
to analyze this data ior of quantum
to extract useful systems based on
information. quantum data.

Quantum algorithms
work directly with
Both the A quantum algo-
Quantum- quantum data to
algorithms and rithm designed
Quantum find patterns or
the data are to learn from
(QQ): insights, making full
purely quantum. quantum states
use of quantum
states. for tasks like op-
timizing quantum
systems or recog-
nizing quantum
patterns.

8
 Data collection:
Capturing Raw data from various sources

Data preparation and cleaning:

Removing incorrect, null and extrane-
ous values. Transforming the data to
the format suitable for the next steps.

Choose a model:
Selecting the correct model to solve the problem.

Training the model:

Adjusting the algorithm paraentes till
the bias and variance balance is achieved.

Parameter tuning
Adjusting the weights of different
parameters to improve the results.

Model evaluation
Testing the trained model with test data
set and check the accuracy of the results
to remove overfitting and underfitting.

Prediction
Actually deploy the model to the real world
for providing answers to the question.

Figure 5: Machine Learning Steps

complexity, robustness to noise, circuit depth to be input to a machine learning algorithm. The
(number of layers),accuracy and optimum bias three approaches discussed are:
variance tradeoff.[24].
The most common approach for implementing
5.2.1 Basis Encoding
QML is the the CQ setting. The two ways to
implement QML is the CQ setting are[47] : In this method the qubits are directly mapped to
the bit values i.e. bit 0 and 1 will be represented
• The translational approach: Some parts of as |0⟩ and |1⟩ respectively. A classical number 110
the classical algorithms are translated i.e. can be represented as the quantum state |110⟩:
run using Quantum approaches. The choice
of using classical or quantum approach is
which approach will give better performance |110⟩ = α |1⟩ ⊗ β |1⟩ ⊗ γ |0⟩ , (2)
(accuracy or speed wise).
where α, β, and γ are the information stored in
• The exploratory approach: The algorithms the form of probability amplitudes for each qubit
are run entirely on quantum systems and state.
may not have any quantum counterpart.
5.2.2 Superposition encoding
5.2 Quantum Data Encoding
As the name suggests, in this encoding scheme,
The work done in [43],[9] discuss the quantum the qubit state is encoded as a superposition of
data encoding methods used to prepare the data the basis states.

9
 Each term is square-normalized by dividing it
r r r by the square root of the sum of squares of all
1 1 1 elements. For the vector X = [1.2, 2.7, 1.1, 0.5],
|110⟩ = |100⟩ + |010⟩ + |001⟩ , (3)
3 3 3 the normalization factor is:
v
u n
uX p √
r
1
r
1
t x2i = 1.22 + 2.72 + 1.12 + 0.52 = 10.19
|78⟩ = |1001110⟩ + |0100111⟩ , (4) i=1
2 2 (9)
The normalized quantum state is:
5.2.3 Angle encoding √ √
1.2 2.7
When the Qubits are represented on the bloch |ψ⟩ = √ |00⟩ + √ |01⟩
sphere, the probabilities of basis states are the 10.19 10.19
√ √
phase shift from the X,Y and Z axes. Thus in 1.1 0.5
+√ |10⟩ + √ |11⟩. (10)
angle encoding, the classical data is represented 10.19 10.19
as a phase shift. Quantum gates use rotation op-
erations around different axes (X, Y, or Z) for 6 Implementation of QML Algorithms
changing the quantum state.
The rotation around the X, Y and Z axis de- In their work [44],the authors have suggested the
noted as Rx (θ),Ry (θ) and Ry (θ) respectively and framework to implement Hybrid Quantum Clas-
is shown in Equations 5,6 and 7. sical algorithms. As discussed previously, Quan-
tum computing can work in tandem with classi-
" # cal system in 4 ways i.e. CC,QC,CQ and QQ.
−iθX/2 cos(θ/2) −i sin(θ/2) Normally, a hybrid quantum algorithm contains
Rx (θ) = e =
−i sin(θ/2) cos(θ/2) 3 parts: the classical part,quantum part and the
(5) interface between the classical and quantum sys-
tems as shown in Fig.6 The data encoding, appli-
" # cation of Oracle, amplitude adjustment and Opti-
−iθY /2 cos(θ/2) − sin(θ/2) mization steps are repeated multiple times in the
Ry (θ) = e = (6)
sin(θ/2) cos(θ/2) range of thousands of iterations to give reliable
" # results.
−iθZ/2 e−iθ/2 0
Rz (θ) = e = (7)
0 eiθ/2 6.1 Quantum Computation
The quantum part comprises of mapping the al-
5.2.4 Amplitude encoding gorithm to an actual circuit with quantum gates.
In this scheme, classical information is encoded The steps involved in this process, shown on the
in the probability amplitudes of quantum state. left side of Fig. 6 in red, are:
If the number of values to be represented is n,
1. Decide the number of qubits required and set
then the number of qubits required are log2 (n) .
them to the basis state of |0⟩.
The values of probabilities of each qubit state is
the square root of the data value divided by the 2. Apply the Hadamard Gate to create a uni-
square root of the sum of squares of all the n data form superposition of all possible states of
values. The quantum state can be represented as: the qubits.
n
X 3. Encode the the resultant states using any of
|ψ(X)⟩ = xi |i⟩ (8) the encoding techniques mentioned in 5.2.
i=1
4. Apply quantum gates to implement the
where xi is the i-th element of the vector X, Quantum Oracle function (Uf ) on the en-
and |i⟩ denotes the computational basis states coded qubits.
of the qubits. For this example, n = 4,
and the coefficients xi correspond to the values 5. The result after applying the Oracle function
[1.2, 2.7, 1.1, 0.5]. is stored in all or few of the Qubits.

10
 Quantum Computation Classical computation

Input No. of Qubits:

Decide the number of Qubits
Data required to solve the problem

Initial State: Set

the initial state
of all Qubits to |0⟩.

Superposition:
Create a superposition of all
the qubit states for such that
each possible state can be
deduced with equal probability.

Data Encoding:
Encode the data to the generated superposition
states using the approaches mentioned in Section. 5.2

Implement the Oracle Uf

Apply the necessary
Qubit gates to imple-
ment the Quantum Oracle.

Store the result Optimization:

The result of the computation Optimize the parameters
is stored in few Qubits. i.e. fine tune the values
iteratively by adjusting
parameters based on the
Amplitude Adjustment objectives of the problem.
The amplitudes of the Qubits
are adjusted (increased and
decreased) to reduce probability
of getting correct results.

Measurement
End
On measurement the quantum values collapses to a bit
values /(0 or 1/) which is measured using classical bits. Result

Figure 6: Framework to implement the Quantum Machine learning.

11
 6. The amplitudes of the undesired states are Table 4: Quantum Development Toolkits
reduced to increase the probability of the
getting the desired result. SDK / Language Hardware Platform
Library / Cloud
7. The final values are measured. As dis- Service
cussed earlier, on measurement, the state of IBM Python IBM Quan- IBM Quan-
a Qubits collapses to classical bit value. The Qiskit tum hard- tum Expe-
binary value will depend on the probability [5] ware rience
of each state. Google Python Google’s Google
Cirq [3] quantum Quantum
hardware AI
6.2 Classical Computation
Ms Q# Supports Azure
In the classical component, classical values ob- QDK multiple Quantum
tained are optimised using classical techniques. [52] hardware
The steps involved, shown on the right hand side providers
of Fig. 6 in cyan, are: through
Azure
1. The final qubit values are measured classi- Rigetti PyQuil Rigetti’s Rigetti
cally.The binary value obtained on measure- Forest supercon- Quantum
ment will depend on the quantum probabil- [50] ducting Cloud
ity of each state. quantum
processors
2. In this step, the values are iteratively modi-
Tensor Python Works on TensorFlow
fied based on the problem statement and ex-
Flow quantum ecosystem
pected results. This is done to increase the
Quan- hardware
probability of correct results. Error correc-
tum [11] using the
tion and mitigation strategies are applied for
Cirq library
better results.
Xanadu Python Xanadu’s Xanadu
Penny- photonic Cloud
6.3 Interface between Quantum and Classical Lane [8] processors
systems D- Python D-Wave Leap Quan-
This phase is shown in Fig. 6 in the violet boxes. Wave quantum tum Cloud
Ocean computers
1. Data Encoding: The binary bits are encoded SDK
using any one of the encoding techniques dis- [53]
cussed in Section 5.2. Open Python Compatible Not
Fermion with Cirq, platform-
2. Measurement: The Qubits are measured [4] Qiskit, specific
where they take a binary value of 0 or 1. and other
frameworks
Amazon Python Multiple Amazon
7 Tools available for Quantum Com- Braket QPUs and Web Ser-
puting [6] simulators vices

Quantum computing has opened doors to solving

the problems earlier considered unsolvable. This
also poses a threat to our existing system security.
All this has sparked a keen interest in Quantum
research. A number of tools are available to de-
velop and run quantum algorithms. Some of the
tools are tabulated in Table 4.

12
8 Conclusion y. URL https://doi.org/10.1038/
s41586-024-08449-y.
Quantum computing is more than just a break- [2] Google Quantum AI. Quantum Com-
through in technology—it’s a whole new way puter | Google Quantum AI — quantu-
of thinking about how we solve problems. By mai.google. https://quantumai.google/
harnessing the strange yet powerful properties quantumcomputer, 2024. [Accessed 11-11-
of quantum mechanics, quantum computers can 2024].
tackle challenges that classical systems simply [3] Google Quantum AI. Cirq Google Quan-
can’t handle. From optimizing complex systems tum AI — quantumai.google. https://
to analyzing massive datasets, the potential im- quantumai.google/cirq, 2024. [Accessed
pact on industries like healthcare, finance, and ar- 27-12-2024].
tificial intelligence is staggering. But this isn’t a [4] Google Quantum AI. OpenFermion Google
technology we’ll see in full force tomorrow. Build- Quantum AI — quantumai.google. https:
ing practical quantum computers is hard. Qubits, //quantumai.google/openfermion, 2024.
the building blocks of quantum systems, are del- [Accessed 27-12-2024].
icate and prone to errors. Scaling up to larger, [5] Gadi Aleksandrowicz, Thomas Alexander,
more reliable systems will take time and inno- Panagiotis Barkoutsos, Luciano Bello, Yael
vation. Even so, companies like IBM and Google Ben-Haim, David Bucher, F Jose Cabrera-
have already made impressive strides with power- Hernández, Jorge Carballo-Franquis, Adrian
ful quantum processors. Meanwhile, cloud plat- Chen, Chun-Fu Chen, et al. Qiskit:
forms like Amazon Braket and Microsoft Azure An open-source framework for quantum
Quantum are making quantum computing tools computing. Accessed on: Mar, 16:
accessible to more people, speeding up the pace 61, 2019. URL "https://zenodo.org/
of discovery and development. records/2562111".
In machine learning, quantum computing of- [6] Amazon Web Services (AWS). Amazon
fers unique ways to improve existing algorithms braket — aws.amazon.com/braket/. https:
and solve new types of problems. Hybrid sys- //aws.amazon.com/braket/, 2024. [Ac-
tems—where quantum and classical computers cessed 12-11-2024].
work together—are showing how these two worlds
[7] Mustafa Bal, Akshay A Murthy, Shaojiang
can complement each other. Libraries and frame-
Zhu, Francesco Crisa, Xinyuan You, Zi-
works, like Qiskit and TensorFlow Quantum, are
wen Huang, Tanay Roy, Jaeyel Lee, David
making it easier for developers and researchers to
van Zanten, Roman Pilipenko, Ivan Nekra-
experiment with quantum ideas in practical ways.
shevich, Andrei Lunin, Daniel Bafia, Yu-
The potential of quantum computing is undeni- lia Krasnikova, Cameron J Kopas, Ella O
able despite the numerous challenges. The future Lachman, Duncan Miller, Josh Y Mu-
of quantum computing will depend not only on tus, Matthew J Reagor, Hilal Cansizoglu,
solving technical problems but also on collabo- Jayss Marshall, David P Pappas, Kim
rations across science, industry, and policymak- Vu, Kameshwar Yadavalli, Jin-Su Oh, Lin
ing. It’s a journey, but one filled with incredi- Zhou, Matthew J Kramer, Florent Lecocq,
ble promise. If progress continues at its current Dominic P Goronzy, Carlos G Torres-
pace, quantum computing could soon transform Castanedo, P Graham Pritchard, Vinayak P
the way we live, work, and think about technol- Dravid, James M Rondinelli, Michael J
ogy. Bedzyk, Mark C Hersam, John Zasadzin-
ski, Jens Koch, James A Sauls, Alexander
Romanenko, and Anna Grassellino. Sys-
References tematic improvements in transmon qubit co-
herence enabled by niobium surface encap-
[1] Rajeev Acharya and et. al Abanin. sulation. Npj Quantum Inf., 10(1), April
Quantum error correction below the 2024. DOI: https://doi.org/10.1038/s41534-
surface code threshold. Nature, 12 024-00840-x.
2024. DOI: 10.1038/s41586-024-08449- [8] Ville Bergholm, Josh Izaac, Maria Schuld,

13
 Christian Gogolin, Shahnawaz Ahmed, dro Davide Ialongo, Massimiliano Pon-
Vishnu Ajith, M Sohaib Alam, Guillermo til, Andrea Rocchetto, Simone Severini,
Alonso-Linaje, B AkashNarayanan, Ali and Leonard Wossnig. Quantum machine
Asadi, et al. Pennylane: Automatic differ- learning: a classical perspective. Pro-
entiation of hybrid quantum-classical com- ceedings of the Royal Society A: Math-
putations. arXiv preprint arXiv:1811.04968, ematical, Physical and Engineering Sci-
2018. DOI: 10.48550/arXiv.1811.04968. ences, 474(2209):20170551, 2018. DOI:
[9] Bhattaraprot Bhabhatsatam and Sucha https://doi.org/10.1098/rspa.2017.0551.
Smanchat. Hybrid quantum encoding: Com- [16] Atom Computing. Home - Atom Com-
bining amplitude and basis encoding for en- puting - atom-computing.com. https://
hanced data storage and processing in quan- atom-computing.com/, 2024. [Accessed 12-
tum computing. In 2023 20th Interna- 11-2024].
tional Joint Conference on Computer Sci- [17] ORCA Computing. Home | ORCA Com-
ence and Software Engineering (JCSSE), puting — orcacomputing.com. https://
pages 512–516, 2023. DOI: 10.1109/JC- orcacomputing.com/, 2024. [Accessed 11-
SSE58229.2023.10201947. 11-2024].
[10] Jacob Biamonte, Peter Wittek, Nicola Pan- [18] QuEra Computing. Quantum Computing
cotti, Patrick Rebentrost, Nathan Wiebe, with Neutral Atoms | QuEra — quera.com.
and Seth Lloyd. Quantum machine learn- https://www.quera.com/, 2024. [Accessed
ing. Nature, 549(7671):195–202, 2017. DOI: 12-11-2024].
https://doi.org/10.1038/nature23474. [19] Rigetti Computing. Rigetti QCS —
[11] Michael Broughton, Guillaume Verdon, qcs.rigetti.com. https://qcs.rigetti.
Trevor McCourt, Antonio J. Martinez, com/qpus, 2024. [Accessed 11-11-2024].
Jae Hyeon Yoo, Sergei V. Isakov, Philip [20] Andrew W. Cross, Graeme Smith, and
Massey, Ramin Halavati, Murphy Yuezhen John A. Smolin. Quantum learning ro-
Niu, Alexander Zlokapa, et al. Tensorflow bust against noise. Phys. Rev. A, 92:
quantum: A software framework for quan- 012327, Jul 2015. DOI: 10.1103/Phys-
tum machine learning. TBD, 03 2020. RevA.92.012327. URL https://link.aps.
[12] Nader H Bshouty and Jeffrey C Jack- org/doi/10.1103/PhysRevA.92.012327.
son. Learning dnf over the uniform [21] Diraq. Diraq — diraq.com. https://diraq.
distribution using a quantum example com/, 2024. [Accessed 11-11-2024].
oracle. In Proceedings of the eighth [22] Vedran Dunjko and Hans J Briegel. Machine
annual conference on Computational learning & artificial intelligence in the quan-
learning theory, pages 118–127, 1995. DOI: tum domain: a review of recent progress. Re-
https://doi.org/10.1137/S0097539795293123. ports on Progress in Physics, 81(7):074001,
[13] Giuseppe Carleo, Ignacio Cirac, Kyle 2018. DOI: https://doi.org/10.1088/1361-
Cranmer, Laurent Daudet, Maria Schuld, 6633/aab406.
Naftali Tishby, Leslie Vogt-Maranto, [23] Vedran Dunjko and Peter Wittek. A non-
and Lenka Zdeborová. Machine learning review of quantum machine learning: trends
and the physical sciences. Reviews of and explorations. Quantum Views, 4:32,
Modern Physics, 91(4):045002, 2019. DOI: 2020. DOI: https://doi.org/10.22331/qv-
https://doi.org/10.1103/RevModPhys.91.045002. 2020-03-17-32.
[14] Simantini Chakraborty, Tamal Das, Saurav [24] Trevor Hastie, Robert Tibshirani, Jerome
Sutradhar, Mrinmoy Das, and Suman Deb. Friedman, Trevor Hastie, Robert Tibshirani,
An analytical review of quantum neural and Jerome Friedman. High-dimensional
network models and relevant research. In problems: p n. The Elements of Statistical
2020 5th International Conference on Com- Learning: Data Mining, Inference, and Pre-
munication and Electronics Systems (IC- diction, pages 649–698, 2009. URL "https:
CES), pages 1395–1400. IEEE, 2020. DOI: //www.sas.upenn.edu/~fdiebold/
10.1109/ICCES48766.2020.9137960. NoHesitations/BookAdvanced.pdf".
[15] Carlo Ciliberto, Mark Herbster, Alessan- [25] Alexander Semenovich Holevo. Bounds for

14
 the quantity of information transmitted by a ibm.com. https://www.ibm.com/topics/
quantum communication channel. Problemy quantum-computing#:~:text=Quantum%
Peredachi Informatsii, 9(3):3–11, 1973. DOI: 20computers%20use%20circuits%20with,
https://doi.org/10.22331/q-2023-04-13-978. individual%20units%20of%20quantum%
[26] Quantum Computing Inc. Quantum 20information, 2024. [Accessed 11-11-
Computing Inc — quantumcomputing- 2024].
inc.com. https://quantumcomputinginc. [35] Abu Kamruzzaman, Yousef Alhwaiti, Av-
com/, 2024. [Accessed 12-11-2024]. ery Leider, and Charles C Tappert. Quan-
[27] Infleqtion. ColdquantaLabs — Infleqtion — tum deep learning neural networks. In
infleqtion.com. https://www.infleqtion. Advances in Information and Communica-
com/coldquantalabs, 2024. [Accessed 12- tion: Proceedings of the 2019 Future of In-
11-2024]. formation and Communication Conference
[28] Oxford Ionics. Oxford Ionics | High (FICC), Volume 2, pages 299–311. Springer,
Performance Quantum Computing — ox- 2020. DOI: https://doi.org/10.1007/978-3-
ionics.com. https://www.oxionics.com/, 030-12385-7_24.
2024. [Accessed 11-11-2024]. [36] Nimish Mishra, Manik Kapil, Hemant
[29] IonQ. IonQ | Trapped Ion Quantum Com- Rakesh, Amit Anand, Nilima Mishra,
puting — ionq.com. https://ionq.com/, Aakash Warke, Soumya Sarkar, San-
2024. [Accessed 11-11-2024]. chayan Dutta, Sabhyata Gupta, Aditya
[30] Abhishek Jadhav, Akhtar Rasool, and Prasad Dash, et al. Quantum ma-
Manasi Gyanchandani. Quantum machine chine learning: A review and current
learning: Scope for real-world prob- status. Data Management, Analytics
lems. Procedia Computer Science, 218: and Innovation: Proceedings of ICD-
2612–2625, 2023. ISSN 1877-0509. DOI: MAI 2020, Volume 2, pages 101–145,
https://doi.org/10.1016/j.procs.2023.01.235. 2021. DOI: https://doi.org/10.1007/978-
URL https://www.sciencedirect. 981-15-5619-7_8.
com/science/article/pii/ [37] Quantum Motion. Quantum Motion - Home-
S1877050923002351. International Con- page — quantummotion.tech. https://
ference on Machine Learning and Data quantummotion.tech/, 2024. [Accessed 11-
Engineering. 11-2024].
[31] Gareth James, Daniela Witten, Trevor [38] Pasqal. Homepage - Pasqal - pasqal.com.
Hastie, Robert Tibshirani, and Jonathan https://www.pasqal.com/, 2024. [Ac-
Taylor. An introduction to statis- cessed 12-11-2024].
tical learning: With applications in [39] PsiQuantum. PsiQuantum-Building
python. Springer Nature, 2023. DOI: the World’s First Useful Quan-
https://doi.org/10.1007/978-3-031-38747-0. tum Computer - psiquantum.com.
[32] Ryan Mandelbaum Jay Gambetta. IBM https://www.psiquantum.com/, 2024.
Quantum delivers on 2022 100x100 perfor- [Accessed 12-11-2024].
mance challenge | IBM Quantum Computing [40] Quantinuum. Quantinuum | Accelerating
Blog — ibm.com. https://www.ibm.com/ Quantum Computing — quantinuum.com.
quantum/blog/qdc-2024, 2024. [Accessed https://www.quantinuum.com/, 2024. [Ac-
14-11-2024]. cessed 11-11-2024].
[33] Jay Gambetta Jerry Chow, Oliver Dial. [41] QuantWare. Accelerating the advent of the
IBM Quantum breaks the 100-qubit quantum computer | QuantWare — quant-
processor barrier | IBM Quan- ware.com. https://www.quantware.com/,
tum Computing Blog — ibm.com. 2024. [Accessed 11-11-2024].
https://www.ibm.com/quantum/blog/ [42] Quobly. Quobly, shaping a better digital
127-qubit-quantum-processor-eagle, future — quobly.io. https://quobly.io/,
2024. [Accessed 11-11-2024]. 2024. [Accessed 11-11-2024].
[34] Ian Smalley Josh Schneider. What [43] Minati Rath and Hema Date. Quan-
Is Quantum Computing? | IBM — tum data encoding: a comparative anal-

15
 ysis of classical-to-quantum mapping tech- tum computing and development with a
niques and their impact on machine learn- high-level dsl. In Proceedings of the
ing accuracy. EPJ Quantum Technol- real world domain specific languages work-
ogy, 11(1), October 2024. ISSN 2196- shop 2018, pages 1–10, 2018. DOI:
0763. DOI: 10.1140/epjqt/s40507-024- https://doi.org/10.1145/3183895.3183901.
00285-3. URL http://dx.doi.org/10. [53] D-Wave Systems. D-Wave Ocean Software
1140/epjqt/s40507-024-00285-3. Documentation; Ocean Documentation 8.1.0
[44] Minati Rath et al. Quantum-assisted simu- documentation — docs.ocean.dwavesys.com.
lation: A framework for designing machine https://docs.ocean.dwavesys.com/en/
learning models in the quantum computing stable/, 2024. [Accessed 27-12-2024].
domain. arXiv preprint arXiv:2311.10363, [54] Prateek Tandon, Stanley Lam, Ben
2023. Shih, Tanay Mehta, Alex Mitev, and
[45] Matthias Rupp, O Anatole Von Lilien- Zhiyang Ong. Quantum robotics: A
feld, and Kieron Burke. Guest ed- primer on current science and future
itorial: Special topic on data-enabled perspectives. Springer, 2017. DOI:
theoretical chemistry. The Journal of https://doi.org/10.2200/S00746ED1V01Y201612QMC01
chemical physics, 148(24), 2018. DOI: [55] Alpine Quantum Technologies. Home - AQT
https://doi.org/10.1063/1.5043213. - Alpine Quantum Technologies — aqt.eu.
[46] Takayuki Sakuma. Application of deep quan- https://www.aqt.eu/, 2024. [Accessed 11-
tum neural networks to finance. arXiv 11-2024].
preprint arXiv:2011.07319, 2020. [56] Peter Wittek. Quantum machine learn-
[47] Maria Schuld and Francesco Petruccione. ing: what quantum computing means to
Supervised Learning with Quantum Comput- data mining. Academic Press, 2014. ISBN
ers. Springer Publishing Company, Incorpo- 9780128009536.
rated, 1st edition, 2018. ISBN 3319964232. [57] Xanadu. Xanadu | Welcome to Xanadu —
xanadu.ai. https://www.xanadu.ai/, 2024.
[48] Maria Schuld, Ilya Sinayskiy, and Francesco
[Accessed 11-11-2024].
Petruccione. An introduction to quan-
[58] Dong Yumin, Mingqiu Wu, and Jinlei
tum machine learning. Contemporary
Zhang. Recognition of pneumonia image
Physics, 56(2):172–185, 2015. DOI:
based on improved quantum neural network.
https://doi.org/10.1080/00107514.2014.964942.
IEEE Access, 8:224500–224512, 2020. DOI:
[49] Peter W Shor. Algorithms for quantum
https://doi.org/10.1109/ACCESS.2020.3044
computation: discrete logarithms and fac-
697.
toring. In Proceedings 35th annual sym-
posium on foundations of computer sci-
ence, pages 124–134. IEEE, 1994. DOI:
https://doi.org/10.1109/SFCS.1994.365700.
[50] Robert S Smith, Michael J Curtis, and
William J Zeng. A practical quan-
tum instruction set architecture. arXiv
preprint arXiv:1608.03355, 2016. DOI:
10.48550/arXiv.1608.03355.
[51] Sandeep Kumar Sood and Pooja. Quan-
tum computing review: A decade of re-
search. IEEE Transactions on Engineer-
ing Management, 71:6662–6676, 2024. DOI:
10.1109/TEM.2023.3284689.
[52] Krysta Svore, Alan Geller, Matthias Troyer,
John Azariah, Christopher Granade, Bet-
tina Heim, Vadym Kliuchnikov, Mariia
Mykhailova, Andres Paz, and Martin
Roetteler. Q# enabling scalable quan-

16