Quantum Computing for Military Applications

Niels M. P. Neumann∗ , Maran P. P. van Heesch∗ , Frank Phillipson∗ and Antoine A. P. Smallegange†
2021 International Conference on Military Communication and Information Systems (ICMCIS) | 978-1-6654-4586-3/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICMCIS52405.2021.9486419

Departments of ∗ ICT Cyber Security and Robustness, † Defense Operations and Human Factors
The Netherlands Organisation for Applied Scientific Research (TNO)
The Hague, The Netherlands
Email: {niels.neumann,maran.vanheesch,frank.phillipson,antoine.smallegange}@tno.nl

Abstract—Quantum computers have the potential to outshine and β such that |α|2 + |β|2 = 1. Using operations we can
classical alternatives in solving specific problems, under the then manipulate this superposition, instead of just the definite
assumption of mature enough hardware. A specific subset of states 0 and 1. Upon measurement of the qubit, only one of
these problems relate to military applications. In this paper
we consider the state-of-the-art of quantum technologies and the two definite states is found with probabilities |α|2 and
different applications of this technology. Additionally, four use- |β|2 , respectively. Another key concept is entanglement, where
cases of quantum computing specific for military applications are two quantum systems are correlated beyond what is possible
presented. These use-cases are directly in line with the 2021 AI classically. Carefully exploiting these two properties allows
strategic agenda of the Netherlands Ministry of Defense. quantum computers to, at least theoretically, break through
Index Terms—Quantum computing, military applications, use
cases, machine learning some of the barriers of classical computing. Yet, quantum
computers are not expected to replace classical computers in
our every-day lives. Quantum computers are for instance not
I. I NTRODUCTION
suitable for writing documents and the simple leisure activities
Computers have become a vital aspect of everyday life and mentioned above. Hence, classical computing devices, such as
are used for a wide-range of applications: from simple leisure laptops and wearables, will remain.
activities to solving some of the most complex problems Two well-known applications of quantum computers are
encountered in society and business. searching unstructured databases [3] and integer factoriza-
We have already mastered how to solve problems such as tion [4, 5]. There are however many more algorithms that
finding the shortest route between two points or multiplying show the power of quantum computers, among which are
numbers efficiently using computers. These problems are often simulating chemical processes [6], optimization problems [7]
time-consuming to do manually, whereas a computer can solve and quantum machine learning algorithms [8, 9].
them faster and often provide a measurably better answer using Where classical computers have been under development for
simple algorithms. There are however also problems for which decades and are therefore quite mature, quantum computers are
there are no such simple algorithms. Examples are pattern still in the early stages of development. Therefore, they are not
recognition, finding anomalies in data and allowing cars to yet used to solve real-life problems. Yet, it is expected that in
drive autonomously. Instead, these problems are often tackled the coming years, quantum computers will be used more often
by providing a computer a model and sufficient data and then and can help in some specific areas. Only recently, a quantum
optimize the model based on the data. That model is then used computer performed a task apparently intractable for classical
to solve the problem. This is the realm of machine learning computers [10]. Note that the task was an artificial one,
and while it is not guaranteed to give the optimal answer, it focused on proving the potential power of quantum computers.
has proven incredibly powerful over the years [1]. There are different approaches to building a quantum com-
However, classical computers cannot be used to solve any puter, each still under active development. Therefore, it is
problem arbitrarily fast. A known barrier is that there are not known which technique will become dominant, neither
problems for which we do not know efficient solutions even is it known if any single technology will become dominant.
with the most powerful classical computers [2]. A well-known Organizations such as IBM [11], D-Wave [12], Microsoft [13],
example, though simple to formulate but difficult to solve, is Rigetti [14], QuTech [15] and many more are participating in
the NP-complete travelling salesmen problem (TSP). Another the world-wide race to build the best quantum computer and
example is integer factorization which is a primer in many some organizations provide access to quantum backends via
currently used asymmetric cryptographic protocols. cloud-based services or one-to-one sales. The computational
Quantum computers operate on a fundamentally different capabilities of these devices are however still limited.
level than classical computers. The elemental computation Additionally to building a quantum computer, it is not yet
units are quantum bits or qubits. Whereas classical bits can an easy task to operate and make use of it. One challenge is
only be in one of two definite states, 0 or 1, qubits can efficiently navigating between different algorithms and hard-
be in complex linear combinations of quantum states, with ware backends, to match operational requirements of specific
slightly different notation: |0i and |1i. A general qubit state use cases. Starting to explore the possibilities now, will help
|ψi is therefore given by |ψi = α |0i + β |1i, for certain α prepare for the quantum future. With timely preparation, the

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
 |0i This specific quantum circuit encodes data points in the
H X
two qubits and can therefore be useful for machine learning
applications. The parameters θi are chosen such that both data
|0i RY (θ1 ) RY (θ2 ) points are encoded in the phases α and β. The first qubit acts
as a label, the second qubit encodes the data points and the
Fig. 1. Amplitude encoding: Two-qubit circuit for data encoding of two final quantum state of this example circuit is given by
normalised two-dimensional data points.
1

|0i |0i → √ |0i |x1 i + |1i |x2 i . (1)
2
quantum potential can directly be used when available and
being the first to be able use it, implies a strategic advantage These quantum circuit diagrams are abstract representations
over those who are not directly able to use quantum computers. of operations on qubits. In practice, operations have to be im-
In this paper we will focus on the potential of quantum plemented on a specific physical quantum hardware backend.
computing in the military domain, an important application Creating these hardware backends is difficult for varying
domain of quantum computing [16]. We first discuss the state reasons, one being that qubits are sensitive and can quickly
of the art quantum technology and the expected technological become useless due to small but destructive interactions with
development in Section II and discuss different applications of the environment. There is no consensus on what the best
quantum computers in Section III. In Section IV we discuss hardware technology is. Some qubit technology examples are
four military use-cases of quantum computing. Three of the superconduction-based transmon qubits [18], qubits based on
use cases focus on machine learning and are directly in line elektron spins [19] and trapped ions [20]. Different vendors
with the 2021 AI strategic agenda of the Netherlands Ministry and universities focus on different technologies, with the state-
of Defense [17]. Conclusions are given in Section V. of-the-art devices having around 70 qubits.
Apart from the variety in quantum hardware, there is also
II. S TATE - OF - THE - ART TECHNOLOGY a variety in how these quantum devices can be programmed.
Different quantum programming languages are ProjectQ [21],
In this section we consider two of the main quantum PyQuil [22], Qiskit [23] and others. Each of these languages
computing initiatives, being gate-based quantum computers however only consider part of the quantum software stack [24]
and quantum annealing devices. Afterwards, we briefly discuss and integration with initiatives for the other parts is vital.
quantum memories and end with a short discussion on how to
scale up the current technology. B. Quantum annealers
Instead of describing quantum computations using circuit
A. Gated quantum computing computing
diagrams, we can also describe them using time-dependent
From the introduction we know that a general qubit state evolving Hamiltonians [25], known as adiabatic quantum
|ψi is a complex linear combination of two definite states |0i computing. If the Hamiltonian evolves slow enough and the
2 2
and |1i: |ψi = α |0i + β |1i, such that |α| + |β| = 1. operation temperature is close to or at absolute zero, the two
Operations on qubits are given by quantum gates that paradigms are equivalent [26]. The solution to the problem
modify |ψi. An example is the Pauli-X gate, which is often corresponds to the lowest energy state of the system.
thought of as the quantum equivalent of the classical NOT In quantum annealing, a heuristic version of adiabatic
gate, interchanging the |0i and |1i state. Quantum gates can quantum computing based on the work of Kadowaki and
also act on multiple qubits simultaneously. An example is the Nishimori [27], the Hamiltonian is evolved fast and often.
CNOT (controlled NOT) gate, which acts on two qubits and The qubits are first brought in an equal superposition of all
applies a Pauli-X gate on the second qubit, conditioned on possible states and then the problem specific Hamiltonian is
the first qubit being in the |1i-state. slowly applied by means of a magnetic field. The qubits inter-
Gate-based quantum computers prepare different qubits in act with this magnetic field and move towards the state with
a known initial state and then apply specific quantum gates on the lowest energy. This can be thought of as a ball rolling down
them to modify the simultaneous quantum state of all qubits a mountain and finally ending up at the valley. Classically, this
to a desired state. Algorithms have to be developed such that process often results in local minima being found. Quantum
the right quantum gates are applied to the right qubits and effects allow for the quantum state to move out of local minima
the final quantum state gives the answer with high probability and end up in the global minima more often. Formulating the
upon measurement. Designing a quantum circuit is however Hamiltonian such that the global minima corresponds to the
a challenging task and suboptimal quantum circuits may limit solution of the problem is challenging.
or even nullify the gains of a quantum computing approach. The current most advanced implementation of quantum
An example two qubit quantum circuit is given in Figure 1. annealers is by D-Wave. These devices require the problem to
The figure should be read from left to right, with each hori- be formulated as Ising Hamiltonians, or in its binary equivalent
zontal line corresponding to a qubit. The blocks correspond to as a QUBO [28]. Both define an optimization problem
specific operations on the qubits and vertical lines correspond
to two qubit controlled operations. QUBO: min xt Qx,
x∈{0,1}n

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
with x ∈ {0, 1}n the decision variables and Q an n × n coeffi- computations. With longer coherence times, more operations
cient matrix. For many combinatorial optimization problems, can be applied on the qubit.
a QUBO formulation is known [28, 29]. Solving a general An issue closely related to the coherence times of qubits
QUBO problem to optimallity is however NP-hard. are imperfect operations. Instead of transforming a qubit to
Similary to how quantum circuit diagrams have to be a known desired state, an imperfect operation transforms a
mapped to gate-based quantum hardware, QUBO formulations qubit to a (unknown) state close to the desired one. Imperfect
have to be mapped to quantum annealing hardware. The D- gates therefore introduce errors, which in turn is related to the
Wave 2000Q version is a quantum annealing device with 2,048 usefulness of the qubit for computations.
qubits, placed in a Chimera-architecture [30]: a 16 × 16 grid To scale up quantum computers, improvements are needed
of unit cells, each unit cell consisting of a bipartite graph of 8 on each of these aspects: More connected qubits are needed
qubits. Every qubit is connected to at most 6 other qubits. The to run algorithms for larger problem instances, and the qubits
most recent version of the device, the D-Wave Advantage, has need to be more coherent. Furthermore, the operations applied
5,000 qubits, placed in a Pegasus architecture [31], containing on the qubits must introduce minimal errors. Quantum error
subgraphs consisting of a complete graph between four qubits correction techniques are one intermediate step in this direc-
or an equal bipartite graph of twelve qubits. Each qubit in this tion [34]. Multiple qubits are encoded in such a way that they
architecture is coupled to fifteen other qubits. together act as a single qubit which is less prone to noise
Different dedicated software modules are available to map and has higher coherent times. This comes at the cost of the
problems to these hardware backends and to decompose prob- number of qubits to be used for computation.
lems that are too large for the hardware in smaller problems Another aspect that can help in scaling up the computational
that can be implemented on the hardware [32]. Newer software power of quantum computers is linking different quantum
modules also allow customability of the decomposition of the computers together, for instance via a quantum internet, and
problems, giving the user more control on the decomposition. distributed operations over the devices [35]. In theory, two
devices of 10 qubits each be be linked together to create a
C. Quantum memory virtual quantum computer of 18 qubits. The remaining two
As we have seen, we can encode data in the phases of qubits are used for interactions between the two devices.
quantum states. If we combine this with entanglement between Local operations can be performed as usual, and the non-
states, each additional qubit allows for a doubling of the num- local operations between qubits on different devices require a
ber of phases we can encode data in. Therefore, we can store shared entangled state between the devices. Here, the two re-
exponentially more data in qubits compared to classical bits. maining qubits are needed. Note that distributing an algorithm
This however presents two of the main challenges quantum introduces another source of errors resulting from the shared
computing faces: loading in data and reading out data. entanglement [36].
Reading out data is a challenge as measurements of quantum
III. T ECHNOLOGY APPLICATION OVERVIEW
states only return a single definite state. Loading in data can
be a similar limiting factor. For example, Grover’s algorithm Even though quantum hardware is still in early stages of
provides a quadratic speed-up for searching for a target item development, there are already many quantum algorithms that
in unstructured databases of size N √. The complexity of a can solve some of the most difficult classical problems. We
quantum algorithm therefore is O( N ), versus a classical now consider different applications of quantum computers.
complexity of O(N ). However, preparing the quantum state First, we discus sampling from probability distributions using
already requires us to consider all database elements, for a total quantum computers in Section III-A and in Section III-B we
number of operations at least equal to the size of the database, discuss quantum approaches to optimization problems. After-
Ω(N ), thereby nullifying the potential quantum gains. wards, we discuss how to use quantum computers for machine
For many quantum algorithms, it is therefore important that learning in Section III-C and for simulating quantum systems
data is loaded efficiently in a quantum state. This can either be in Section III-D. In Section III-E we discuss the implications
by using output quantum data of another process as input for of quantum computers on certain types of cryptography.
the quantum algorithm, or by efficiently querying data from a
memory or quantum RAM [33]. Note that quantum RAM is A. Sampling from probability distributions
out of reach for near-to medium-term applications. Sampling from probability distributions is important and has
a wide range of applications, e.g., determining which lots to
D. How to scale up the current technology search for illegal activity in harbors, as searching all possible
Two important limiting factors in current state-of-the-art locations is often too time-consuming. Random sampling of
quantum technology are increasing the number of connected protocol parameters is also used in many security protocols.
qubits and the coherence time of them. Creating additional Not all probability distributions are however easy to sample
qubits is not that hard, however, creating them such that they from. Some quantum mechanical processes can efficiently pro-
can interact with other qubits is hard. The coherence time of a duce probability distributions that which are hard to replicate
qubit relates to how long a qubit remains sufficiently useful for efficiently classically, for instance due to entanglement [37,

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
38]. Quantum computers can be used to generate such proba- networks show similarity with the classical variants, often
bility distributions efficiently and sample from them by mea- with gate or circuit parameters being optimized [8]. Various
surements. Interestingly, dedicated devices, focused on specific quantum neural network approaches have been proposed, the
problem, can already outperform classical devices for specific gains of which can be grouped in three classes [9, 47, 48]:
probability distributions [10, 39, 40]. Quantum computers are 1) Run-time improvements: the total running time of the
already used to sample from specific probability distributions, algorithm is reduced, for instance due to faster sampling
however with limited practical applicability. techniques [49, 50];
2) Learning efficiency improvements: less training samples
B. Optimization algorithms
or training iterations are needed to obtain similar per-
Optimization algorithms are among the first to be run formance [51];
on near-term hardware and already various optimization al- 3) Capacity improvements: increased (storage) capacity of
gorithms exist for both gate-based quantum computing and associative or content-addressable networks [52, 53].
quantum annealing. The first hybrid quantum-classical implementations have al-
The most common optimization algorithms for gate-based ready been implemented for use-cases. We expect that within
quantum computing are the quantum approximate optimization five years hybrid quantum-classical machine learning algo-
algorithm (QAOA) [41] and variational quantum eigensolvers rithms will be used for solving multiple real-life problems.
(VQEs) [42]. In both cases, based on the problem formulation,
specific operations are implemented, the parameters of which D. Quantum simulation
are optimized in a hybrid quantum-classical fashion. A chal- Simulating nature at its most elementary level, i.e., sim-
lenge for both algorithms is the initial choice of parameters. ulating quantum systems, is a complex task. It was also
A suboptimal choice may result in the optimization to get this task that was one of the main arguments to work on
stuck in local optima. There are still questions regarding how developing quantum computers [54]. With each additional
well these algorithms will perform for practical problems. We qubit, the required classical resources double, making it hard
expect that within five years, these algorithms can be used for to simulate all but the most simple quantum systems. Quantum
small but practical use-cases. computing offers an efficient way of simulating complex
Using currently available quantum annealing hardware opti- quantum systems [6, 55]. This in turn can help in developing
mization problems can be solved that are formulated as QUBO new materials or personalized medicines. Quantum computers
problems. Even though the solutions are not guaranteed to be can already offer gains for small quantum systems with a few
optimal, depending on the specific problem, close to optimal years. It will still be years before quantum computers can help
solutions can be found quickly. This approach is already used simulating large and complex molecule and proteins.
for various problems such as scheduling of assets [43], vehicle
routing [7] and optimizing stock portfolios [44]. For vehicle E. Cryptography
routing for instance, results were found in real-time and Two of the most influential algorithms from the early days
congestion measurably decreased. This approach is already of quantum computing are the algorithms by Shor [4, 5] and
used for small but practical use-cases. Grover [3]. The former can be used to break certain asymmet-
ric cryptographic protocols, including RSA, Diffie-Hellman
C. Machine learning and elliptic curve cryptography, whereas Grover’s algorithm
Different algorithms and models fall under the concept has the potential to weaken certain symmetric protocols such
of machine learning. For each, data is used to learn some as AES and the SHA-familiy of hash functions.
desired outcome. Each of the models has its own gains and Shor’s algorithm gives an exponential speed-up over clas-
benefits. Neural networks form a special class of machine sical algorithms by solving the discrete-log problem, the
learning algorithms. Quantum computing can provide benefits primer of the Diffie-Hellman protocol. For an integer N to
for both machine learning in general and neural networks factor, Shor’s algorithm computes in superposition the values
specifically. Quantum variants of support vector machines g x mod N for a fixed integer g and varying integers x. A
can for instance provide an exponential computational im- quantum Fourier transform then returns information about the
provement over classical variants [45]. Another example is period of x of this function, from which the prime factors can
nearest-neighbor classification, where a worst-case polynomial be determined and with which the protocols are broken.
improvement is found [46] and for specific cases even an Grover’s algorithm provides a quadratic speed-up in break-
exponential improvement over classical alternatives. ing symmetric encryption protocols such as AES. Grover’s
Neural networks are inspired by nature, information is algorithm is a search algorithm and can be used to find pre-
passed between neurons that take either discrete or continuous images quadratically faster than classical algorithms.
values. Different neurons are connected by edges with certain These algorithms have significant implications on currently
weights and nodes are updated based on values of connected used encryption, given sufficiently strong quantum hardware.
neurons, the corresponding weights and an updating rule. Different implementations have been proposed to implement
Often the neurons are grouped in layers with connections only these algorithms, each with specific benefits [56, 57, 58].
existing between nodes of consecutive layers. Quantum neural Depending on the implementation, at least 2,000 error-free

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
qubits are needed to run these algorithms. We expect that it B. Improving radar and sonar data processing
will be more than fifteen years before these algorithms can be
Situational awareness is the start for every (military) deci-
used for breaking encryption used in practical settings.
sion. Improving situational awareness directly leads to better
decisions and success on the battlefield. Obtaining information
IV. M ILITARY USE CASES superiority is therefore paramount. Data obtained from sensors
helps in understanding the environment and hence in obtaining
In the previous sections we considered various applications information superiority. Correctly interpreting the obtained
of quantum computers and the state of the art of both gate- data is vital when using it in a sensible way.
based quantum computers and quantum annealers. In this
We foresee a number of problem areas within data process-
section we consider different military related use-cases of
ing where quantum computers can play an important role.
quantum computing, in line with various strategic documents
The first is optimizing the waveshape form in radar re-
of the Netherlands Defense Organisation. Quantum technol-
search [63]. The design of the waveform is an important aspect
ogy, including quantum computing, is one of the key enabling
of multi-input, multi-output (MIMO) radar research as the
technologies identified by the Netherlands Ministry of Defense
performance of MIMO radar depends on the specific signal
in their most recent Strategic Knowledge and Innovation
design. This is a challenging and topical problem in spectrally
Agenda [59], and the Netherlands Defense Strategy on Data
crowded environments, especially due to the increasing de-
Science and AI [17]. This strategy defines long term (2035)
mand of both military and civilian wireless services. The range
ambitions on five application area’s: unmanned autonomous
resolution and the target detection capability can be improved
systems, military decision support & intelligence, business
by decreasing the width of the probing pulse and increasing the
operations, and company safety. For each application area
transmitted energy. This however also leads to larger transmit
additional short term (2025) goals are defined. To achieve these
peaks, which are generally unacceptable. Quantum annealing
ambitions, an integrated approach to technology, governance,
can help in optimizing the design of the transmit code and the
capacity and organizational change is required. Furthermore,
target receive filters, such that target capabilities are as high as
objectives are defined that give direction to activities needed
possible, while transmit peaks power levels remain acceptable.
to reach the defined ambitions. In this section we will indicate
Reconstructing an image from discrete observations or
how quantum computing can help in reaching these ambitions.
signals is notoriously difficult [64, 65]. The technique com-
Three of the four use cases presented here are directly
pressive sensing can be used to acquire and reconstruct signals
related to the Netherlands Defense AI strategy. Our first use-
more efficiently, by finding sparse solutions to underdeter-
case in Section IV-A considers efficiently scanning footage
mined linear systems. Quantum annealers already find reason-
of enemy territory. In Section IV-B we consider the gains
able results for toy-problems. Gate-based quantum computers
quantum computing offers in improved data processing to
show similar potential, however, not for near-term devices.
benefit ones information superiority. Determining the best
A third problem area is normalization of data, using the av-
course of action and how to improve on it is discussed in
erage or median value. Finding these values however requires
Section IV-C. Implications of quantum computing on currently
us to process the whole database. Instead, using a variant of
used cryptography are discussed in Section IV-D.
Grover’s algorithm, both can be computed with quadratically
less steps than classical approaches require.
A. Recognizing hostile vehicle in drone footage
Unmanned autonomous systems such as drones can be used C. Improved course of action determination
to safely and efficiently scan and assess hard-to-access or Military aim to improve their observe–orient–decide–act
high-risk areas. Efficiently scanning the thus obtained footage (OODA) loop to be faster than that of adversaries. Most
is important as it provides information that can make the developments aim to improve the first two aspects (observe
difference between the success and failure of a mission. and orient). This is also reflected in current military support
Furthermore, there is a time-constraint where decisions have tools, with support on for example threat evaluation and
to be made fast before information is outdated again. Even weapon assignment. The holy grail lies in complete decision
though classical algorithms have proven powerful over the last support where military have situational understanding and are
years, they can also be fooled relatively simple [60]. supported by system generated course of action and valida-
Quantum computing allows for new enhanced pattern recog- tion. With the rising possibilities of machine learning and
nition algorithms with the potential to outperform classical al- simulation in combination with improved computing power,
gorithms [61, 62]. Another drawback of recognition algorithms true course of action generation and validation becomes within
is often the lack of training data, the generation of which is reach. At this stage however, the shear infinite combinations of
expensive. One way of generating new data samples is by own and adversary actions allows only for very narrow defined
creating efficient data representations of actual data samples actions, such as optimal route finding or fire distribution.
and generate new samples from this efficient representation. Combining these and adding extra constraints or alternate
Both gate-based and quantum annealing-based approach exist actions such as communication limitations, supplies or multi
for this task of efficient data generation [50, 51]. domain support, significantly increases the complexity.

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
 One approach to these complex environments is reinforce- Currently, two main quantum computing paradigms exist.
ment learning, where actions by agents result in a (possibly The first is gate-based quantum computing, which can run all
negative) reward. Based on this reward, the agents might take quantum algorithms, but is harder to build. The second are
different actions in future iterations. After sufficient training, quantum annealers, heuristic versions of adiabatic quantum
the reinforcement learning model can be used to determine computers. On current quantum annealing hardware only a
optimal course of action in unknown environments. specific type of optimization problems can be solved. For
Using quantum annealers, quantum reinforcement Boltz- many problems such a formulation exists however. As quan-
mann machines can provide results faster compared to classical tum annealers are in general easier to build and scale than
alternatives [66, 67]. The quantum algorithm learns using less gate-based devices, the number of qubits these devices offer
training iterations. It is to be noted that the performance of the is higher. Using these devices run-time improvements can
classical algorithm catches up with the quantum performance already be achieved for larger problem instances.
given sufficient learning iterations. Similar algorithms also We presented four use-cases of quantum computing specif-
exist for gate-based quantum computers, where three registers ically focused on military applications. Three of these use-
are defined, one for the environment, one for the agents, and cases are directly related to the Netherlands Defense strategy
one for the interaction between the two [68]. on Data Science and AI.
The gains of quantum computing over classical computing
D. Decrypting communication are still largely theoretical. Over time new applications or
algorithms will be found that are currently unknown. It is
Most of today’s information is kept secure using crypto-
however important to already start preparing for the quantum
graphic protocols, to ensure confidentiality, integrity, and au-
future in order to be fully prepared and ready to directly use the
thenticity of the information. Especially in mission critical sit-
capabilities of quantum computers once they become available.
uations it is extremely important that communication is secure.
Knowledge on enemy communications can therefore have far- R EFERENCES
reaching consequences, demonstrated already 75 years ago as [1]D. Silver et al. “Mastering the game of Go with deep
it was the breaking of the Enigma cipher that played a large neural networks and tree search”. In: Nature 529.7587
role in the victory of the allies in World War II. (Jan. 2016), pp. 484–489.
Many of today’s encryption protocols rely on computa- [2] S. Aaronson. “The Limits of Quantum Computers”. In:
tional assumptions, i.e., rely on some problems being hard Scientific American 298.3 (Mar. 2008), pp. 62–69.
to solve using classical computers without explicit knowledge [3] L. K. Grover. “A Fast Quantum Mechanical Algorithm
of the key. As mentioned in Section III-E, a sufficiently for Database Search”. In: Proceedings of the Twenty-
strong quantum computer can be used to weaken or even Eighth Annual ACM Symposium on Theory of Comput-
break enemies current encryption and consequently read their ing. STOC ’96. Philadelphia, Pennsylvania, USA: As-
communication. To fully benefit from this scenario one can sociation for Computing Machinery, 1996, pp. 212–219.
already store encrypted communications now, and wait until [4] P. W. Shor. “Algorithms for quantum computation:
quantum computers are advanced enough to break today’s discrete logarithms and factoring”. In: Proceedings 35th
security measures. This principle is also called the ‘store now, Annual Symposium on Foundations of Computer Sci-
decrypt later’ paradigm. We expect that it will take at least ence. IEEE Comput. Soc. Press, 1994, pp. 124–134.
fifteen more years before quantum computers are sufficiently [5] P. W. Shor. “Polynomial-Time Algorithms for Prime
powerful for these applications. Factorization and Discrete Logarithms on a Quantum
Note that there are classical cryptographic protocols for Computer”. In: SIAM J. Comput. 26.5 (1997), pp. 1484–
which there is no known efficient classical or quantum algo- 1509.
rithm that one can exploit to break the protocols, called post- [6] I. Kassal et al. “Simulating Chemistry Using Quantum
quantum cryptography. Additionally, quantum effects can be Computers”. In: Annual Review of Physical Chemistry
used to assure unconditionally secure encryption for specific 62.1 (May 2011), pp. 185–207.
applications [69]. Migration to use quantum-safe cryptography [7] F. Neukart et al. “Traffic Flow Optimization Using a
is therefore advised to start sooner rather than later [70]. Quantum Annealer”. In: Frontiers in ICT 4 (2017),
p. 29.
V. C ONCLUSION [8] M. Schuld, I. Sinayskiy, and F. Petruccione. “An intro-
Quantum computing has the potential to greatly outshine duction to quantum machine learning”. In: Contempo-
classical computers in specific tasks. For some applications, rary Physics 56.2 (Oct. 2014), pp. 172–185.
near-term quantum computers are not powerful enough and it [9] N. M. P. Neumann, F. Phillipson, and R. Versluis.
will still be years before they are used for solving practical “Machine learning in the quantum era”. In: Digitale
use-cases. For other applications, already on nearer terms Welt 3.2 (Apr. 2019), pp. 24–29.
advantages over classical devices can be reached. Applications [10] F. Arute et al. “Quantum supremacy using a pro-
where quantum computing already is helpful on the near-term grammable superconducting processor”. In: Nature
are often related to sampling, optimization and simulation. 574.7779 (Oct. 2019), pp. 505–510.

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
[11] IBM. IBM Quantum Experience. 2018. URL: https : / / [30] C. C. McGeoch. “Adiabatic quantum computation and
quantum-computing.ibm.com/ (visited on 03/03/2020). quantum annealing: Theory and practice”. In: Synthesis
[12] D-Wave. D-Wave Leap. 2019. URL: https : / / cloud . Lectures on Quantum Computing 5.2 (2014), pp. 1–93.
dwavesys . com / leap / login / ?next = /leap/ (visited on [31] D-Wave-Systems. The D-Wave Advantage System: An
03/03/2020). Overview. 2020.
[13] Microsoft. Azure Quantum. 2021. URL: https://azure. [32] M. Booth, S. P. Reinhardt, and A. Roy. Partitioning
microsoft . com / en - us / services / quantum/ (visited on Optimization Problems for Hybrid Classical/Quantum
03/03/2020). Execution. Tech. rep. D-Wave Systems, Sept. 2017.
[14] P. J. Karalekas et al. “A quantum-classical cloud plat- [33] V. Giovannetti, S. Lloyd, and L. Maccone. “Quantum
form optimized for variational hybrid algorithms”. In: Random Access Memory”. In: Phys. Rev. Lett. 100 (16
Quantum Science and Technology 5.2 (Apr. 2020), Apr. 2008), p. 160501.
p. 024003. [34] D. Gottesman. An Introduction to Quantum Error
[15] QuTech. Quantum Inspire. 2020. URL: https : / / www. Correction and Fault-Tolerant Quantum Computation.
quantum-inspire.com/ (visited on 03/03/2020). 2009. eprint: arXiv:0904.2557.
[16] M. Piattini et al. “The Talavera Manifesto for Quantum [35] A. Yimsiriwattana and S. J. Lomonaco. “Generalized
Software Engineering and Programming.” In: QAN- GHZ states and distributed quantum computing”. In:
SWER. 2020, pp. 1–5. American Mathematical Society (2005), pp. 131–147.
[17] Ministry of Defense. Defensie Strategie Data Science [36] N. M. P. Neumann, R. van Houte, and T. Attema.
en AI 2020-2025 - Werken aan een slimme krijgsmacht. “Imperfect Distributed Quantum Phase Estimation”. In:
2021. Computational Science – ICCS 2020. Ed. by V. V.
[18] J. Koch et al. “Charge-insensitive qubit design derived Krzhizhanovskaya et al. Cham: Springer International
from the Cooper pair box”. In: Phys. Rev. A 76 (4 Oct. Publishing, 2020, pp. 605–615.
2007), p. 042319. [37] J. S. Bell. “On the Einstein Podolsky Rosen paradox”.
[19] D. Loss and D. P. DiVincenzo. “Quantum computation In: Physics Physique Fizika 1 (3 Jan. 1964), pp. 195–
with quantum dots”. In: Phys. Rev. A 57 (1 Jan. 1998), 200.
pp. 120–126. [38] J. F. Clauser et al. “Proposed Experiment to Test Local
[20] R. Blatt and D. Wineland. “Entangled states of Hidden-Variable Theories”. In: Phys. Rev. Lett. 23 (15
trapped atomic ions”. In: Nature 453.7198 (June 2008), Oct. 1969), pp. 880–884.
pp. 1008–1015. [39] S. Aaronson and A. Arkhipov. “The Computational
[21] D. S. Steiger, T. Häner, and M. Troyer. “ProjectQ: an Complexity of Linear Optics”. In: Theory of Computing
open source software framework for quantum comput- 9.1 (2013), pp. 143–252.
ing”. In: Quantum 2 (Jan. 2018), p. 49. [40] H. Goto, Z. Lin, and Y. Nakamura. “Boltzmann sam-
[22] R. S. Smith, M. J. Curtis, and W. J. Zeng. A Practi- pling from the Ising model using quantum heating of
cal Quantum Instruction Set Architecture. 2016. arXiv: coupled nonlinear oscillators”. In: Scientific Reports 8.1
1608.03355 [quant-ph]. (May 2018).
[23] H. Abraham et al. Qiskit: An Open-source Framework [41] E. Farhi, J. Goldstone, and S. Gutmann. A Quantum Ap-
for Quantum Computing. 2019. proximate Optimization Algorithm. 2014. eprint: arXiv:
[24] R. F. van den Brink, N. M. P. Neumann, and F. 1411.4028.
Phillipson. “Vision on Next Level Quantum Software [42] N. Moll et al. “Quantum optimization using variational
Tooling”. In: 11th International Conference on Future algorithms on near-term quantum devices”. In: Quantum
Computational Technologies and Applications. May Science and Technology 3.3 (June 2018), p. 030503.
2019, pp. 16–23. [43] K. Ikeda, Y. Nakamura, and T. S. Humble. “Application
[25] E. Farhi et al. Quantum Computation by Adiabatic of Quantum Annealing to Nurse Scheduling Problem”.
Evolution. 2000. eprint: arXiv:quant-ph/0001106. In: Scientific Reports 9.1 (Sept. 2019).
[26] D. Aharonov et al. “Adiabatic Quantum Computation [44] G. Rosenberg et al. “Solving the Optimal Trading
is Equivalent to Standard Quantum Computation”. In: Trajectory Problem Using a Quantum Annealer”. In:
SIAM J. Comput. 37.1 (Apr. 2007), pp. 166–194. IEEE Journal of Selected Topics in Signal Processing
[27] T. Kadowaki and H. Nishimori. “Quantum annealing in 10.6 (2016), pp. 1053–1060.
the transverse Ising model”. In: Physical Review E 58.5 [45] P. Rebentrost, M. Mohseni, and S. Lloyd. “Quantum
(1998), p. 5355. Support Vector Machine for Big Data Classification”.
[28] F. Glover, G. Kochenberger, and Y. Du. “A tutorial on In: Phys. Rev. Lett. 113 (13 Sept. 2014), p. 130503.
formulating and using qubo models”. In: arXiv preprint [46] N. Wiebe, A. Kapoor, and K. M. Svore. “Quantum
arXiv:1811.11538 (2018). Nearest-Neighbor Algorithms for Machine Learning”.
[29] A. Lucas. “Ising formulations of many NP problems”. In: Quantum Information and Computation 15.3-4 (Mar.
In: Frontiers in Physics 2 (2014), p. 5. 2015), pp. 318–358.

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.
[47] V. Dunjko and H. J. Briegel. “Machine learning & [63] P. Stoica, J. Li, and M. Xue. “Transmit codes and
artificial intelligence in the quantum domain: a review receive filters for radar”. In: IEEE Signal Processing
of recent progress”. In: Reports on Progress in Physics Magazine 25.6 (2008), pp. 94–109.
81.7 (June 2018), p. 074001. [64] J. H. G. Ender. “On compressive sensing applied to
[48] F. Phillipson. “Quantum Machine Learning: Benefits radar”. In: Signal Processing 90.5 (2010), pp. 1402–
and Practical Examples.” In: QANSWER. 2020, pp. 51– 1414.
56. [65] T. Guerneve, K. Subr, and Y. Petillot. “Three-
[49] A. Perdomo-Ortiz et al. “Opportunities and challenges dimensional reconstruction of underwater objects using
for quantum-assisted machine learning in near-term wide-aperture imaging SONAR”. In: Journal of Field
quantum computers”. In: Quantum Science and Tech- Robotics 35.6 (2018), pp. 890–905.
nology 3.3 (June 2018), p. 030502. [66] D. Crawford et al. “Reinforcement Learning Using
[50] M. Benedetti, J. Realpe-Gómez, and A. Perdomo- Quantum Boltzmann Machines”. In: Quantum Info.
Ortiz. “Quantum-assisted Helmholtz machines: A quan- Comput. 18.1–2 (Feb. 2018), pp. 51–74.
tum–classical deep learning framework for industrial [67] N. M. P. Neumann et al. “Multi-agent Reinforcement
datasets in near-term devices”. In: Quantum Science and Learning Using Simulated Quantum Annealing”. In:
Technology 3.3 (May 2018), p. 034007. Computational Science – ICCS 2020. Ed. by V. V.
[51] T. J. van Dam et al. “Hybrid Helmholtz machines: a Krzhizhanovskaya et al. Cham: Springer International
gate-based quantum circuit implementation”. In: Quan- Publishing, 2020, pp. 562–575.
tum Information Processing 19.6 (June 2020). [68] V. Dunjko, J. M. Taylor, and H. J. Briegel. “Advances in
[52] P. Rebentrost et al. “Quantum Hopfield neural network”. quantum reinforcement learning”. In: 2017 IEEE Inter-
In: Phys. Rev. A 98 (4 Oct. 2018), p. 042308. national Conference on Systems, Man, and Cybernetics
[53] N. Meinhardt, N. M. P. Neumann, and F. Phillipson. (SMC). 2017, pp. 282–287.
“Quantum Hopfield Neural Networks: A New Approach [69] N. M. P. Neumann, M. P. P. van Heesch, and P. de Graaf.
and Its Storage Capacity”. In: Computational Science – “Quantum Communication for Military Applications”.
ICCS 2020. Ed. by V. V. Krzhizhanovskaya et al. Cham: In: Proceedings of the 25th ICCRTS conference (2020).
Springer International Publishing, 2020, pp. 576–590. [70] F. Muller and M. van Heesch. Migration to Quantum-
[54] R. P. Feynman. “Simulating physics with computers”. Safe Cryptography. Tech. rep. 2020.
In: International Journal of Theoretical Physics 21.6-7
(June 1982), pp. 467–488.
[55] B. P. Lanyon et al. “Towards quantum chemistry on
a quantum computer”. In: Nature Chemistry 2.2 (Jan.
2010), pp. 106–111.
[56] S. Beauregard. “Circuit for Shor’s algorithm using 2n+3
qubits”. In: Quantum Inf. Comput. 3.2 (2003), pp. 175–
185.
[57] T. Häner, M. Roetteler, and K. M. Svore. “Factoring
Using 2n + 2 Qubits with Toffoli Based Modular Mul-
tiplication”. In: Quantum Info. Comput. 17.7–8 (June
2017), pp. 673–684.
[58] M. Grassl et al. “Applying Grover’s algorithm to AES:
quantum resource estimates”. In: Proceedings of the 7th
International Conference on Post-Quantum Cryptogra-
phy (PQCrypto’16), Fukuoka, Japan. Vol. 9606. Lecture
Notes in Computer Science. See also arXiv preprint
arXiv:1512.04965. Springer, Sept. 2016, pp. 29–43.
[59] Ministry of Defense. Strategische Kennis- en Inno-
vatieagenda 2021 2025. 2020.
[60] A. Adhikari et al. Adversarial Patch Camouflage against
Aerial Detection. 2020. eprint: arXiv:2008.13671.
[61] H. Neven et al. Training a Large Scale Classifier with
the Quantum Adiabatic Algorithm. 2009. eprint: arXiv:
0912.0779.
[62] Y. Li et al. “A quantum deep convolutional neural
network for image recognition”. In: Quantum Science
and Technology 5.4 (July 2020), p. 044003.

Authorized licensed use limited to: Univ Complutense de Madrid. Downloaded on April 08,2025 at 16:16:17 UTC from IEEE Xplore. Restrictions apply.