Quantum Artificial Intelligence in Financial Crime
“We know from history that we just don’t have the imagination to anticipate where new information technologies can carry us.” — John Preskell
This article introduces quantum artificial intelligence (QAI) and its substantial impact on the financial crime domain because it might solve some problems considerably faster than the best-known classical machine learning algorithms. QAI focuses on developing quantum algorithms for enhancing computational tasks within artificial intelligence (AI), including the machine learning subdomain. Quantum phenomena such as superposition, interference, entanglement, and tunneling allow Quantum Computing (QC) to perform computations that are much more efficient than classical AI algorithms used in the financial crime domain. Artificial intelligence and quantum computing are bound to evolve together to challenge the classical computing paradigms. This article focuses on fundamental principles of QAI, and its various applications within financial crime services-based AI systems. Harnessing the power of a quantum computer to bring business advantage is a journey that will take time; users tempted to take a ‘wait-and-see’ strategy could find themselves years behind the competition. With the materialization of quantum computers and projected growth over the next decade, it is expected that early adopters will achieve breakthroughs that enable new business models. Hence, it is advantageous for organizations to act now and begin understanding quantum computing’s possible use cases and begin putting a plan in place.
Quantum Artificial Intelligence (QAI) integrates knowledge and methods from multidisciplinary fields, using a synthesis of approaches focusing on quantum computing for boosting complex computational tasks within artificial intelligence and its subdomains as machine learning.
Quantum computing (QC) and artificial intelligence (AI) are transformational technologies as artificial intelligence is likely to require quantum computing to achieve significant progress. While the influence of quantum computing on artificial intelligence technologies is pretty obvious, not many think about the opposite relationship. The bilateral affinity between quantum computing and artificial intelligence maintains that by applying the quantum supremacy argument to artificial intelligence, the genesis of quantum computing will make possible the foundation of new paradigms of artificial neural networks that are impossible today. On the other hand, recent developments in artificial intelligence are also influencing the evolution of quantum technologies.
Quantum Computing can provide a computation boost to artificial intelligence, enabling it to address more complex problems in Artificial General Intelligence AGI (a branch of computer science involved in building intelligent machines capable of performing human-like tasks).
Both quantum computing and artificial intelligence are very active fields with an overwhelming speed of new developments in the last few years, therein are vast possibilities for interactions. Quantum algorithms and artificial intelligence might irrefutably constitute an even deeper connection due to mechanics phenomena. The emerging field of QAI uses quantum-inspired algorithms to solve computation tasks related to AI. This combination may be highly synergetic because advances in AI are becoming more demanding in computational resources. This course is outgrowing the ongoing increase in the availability of computing power by a large margin.
However, fully scaled quantum technology is still a way off, but some financial institutions already anticipate the potential value. Financial crime services require the ability to assess a range of possible outcomes. To do this, banks use machine learning algorithms to calculate statistical probabilities. In a reality, where data are abundant, ever-more-powerful computers are essential to estimating precisely probabilities, risks, correlations, and causations. Quantum technology is approaching the mainstream, and in this regard, several banks are turning to a new generation of processors that leverage the principles of quantum physics to process vast volumes of data at superfast speed.
Google, a leader in the field, said in 2019 that its Sycamore quantum processor took a little more than three minutes to perform a task that would occupy a supercomputer for thousands of years. Google AI describes quantum computing as “a new paradigm that will play a big role in accelerating tasks for AI. We want to offer researchers and developers access to open source frameworks and computing power that can operate beyond classical capabilities.” Google has open-source frameworks which explicitly designed to develop novel quantum algorithms to help solve near-term applications for practical problems. One tool is Cirq, and another is OpenFermion.
Goldman Sachs recently announced that they could introduce quantum algorithms to price financial instruments in as soon as five years . Honeywell, a multinational conglomerate company that operates in aerospace and building technologies prognosticates that quantum computing will form a $1 trillion industry in the decades ahead . JPMorgan also experimented with Honeywell’s quantum computer to ease mathematical operations that involve Fibonacci numbers . JPMorgan and Citigroup, meanwhile, have set up quantum computing initiatives and even bought stakes in computing start-ups .
In late 2019, Wells Fargo joined the IBM Q program, a community of companies, startups, academic institutions, and research labs working to explore practical applications . European banks are also exploring quantum computing opportunities. BBVA has formed a partnership to explore portfolio optimization and more efficient Monte Carlo modeling . Also in Spain, Caixa Bank is running a trial hybrid framework of quantum and conventional computing with the aim of better-classifying credit risk profiles .
In late 2019, a Bank of America strategist said quantum computing would be “as revolutionary in the 2020’s as smartphones were in the 2010s.” . However, from a business line perspective, the most promising use cases are likely to be those that require highly intricate, compound, and/or exceptionally fast models.
These initiatives make sense because they allow financial institutions to test quantum algorithms on simulators or the cloud without acquiring full-scale quantum computers. It appears to be a sensible strategy as long as quantum computers remain subcritical for practical applications and there is no dominant design for scaling quantum capabilities.
The aptitude for calculating rapidly and precisely optimal risk scores for fraudulent transactions establishes a significant competitive advantage. Accurate estimates of historical behavioral patterns should lead to better optimization within decision-making systems. In a broad sense, fraud and money laundering solutions across a range of corporate finance activities can be improved by insights into the size, high-dimensionality, sparsity, complexity of relationships between entities, and materiality of risks. Payments and money transfers can be secure through better encryption as well.
Quantum computing applications might bring many advancements to non-convex optimization and constrained models within the financial crime domain by attracting initial investors and users. Quantum computing may deliver specific benefits like enhancement in customer engagement using behavioral data, greater compliance, and quicker response to market volatility.
The quantum computer solution space is orders of magnitude more enormous even than supercomputer systems. Double traditional computers’ capability, approximately twice the number of transistors are needed. In a quantum computer, computational power doubled whenever a qubit is added. It may take several years for quantum computing to attain wide commercial applications. However, many experts predict it would lead to the breakthrough of certain products and services within different business domains in general and in fighting financial crimes solutions in particular. Financial institutions may use quantum computing to restructure their advanced analytics solutions to prevent fraud.
Quantum Computing — Fundamental Principles
The Quantum computing process concentrates on creating systems and technology-based quantum mechanics phenomena such as superposition, entanglement, interference, and tunneling that allow performing computations that are much more efficient than classical AI algorithms used in financial crime services.
Quantum computing uses quantum bits (qubits) which are more advanced than the traditional bits used by a classical computer. According to the nature of quantum mechanics law, qubits operate differently than bits at a subatomic level. Under quantum entanglement, scientists can push multiple qubits into the same state, even if the qubits aren’t in contact with each other. 2-qubits represent four states (00, 01, 10, 11). Due to the superposition phenomenon, the qubits can represent all four states at the same time. The superposition of two qubits increases exponentially as more qubits entangle with each other. Thereby, a two-qubit structure contains four possible values, a twenty-qubit structure contains more than a million.
A Quantum logic gate (or quantum gate) is a rudimentary quantum circuit operating on a small number of qubits. They are the building blocks of quantum circuits like classical logic gates are for conventional digital circuits. Classical gates operate on classical bits, while quantum gates operate on quantum qubits. Unlike many classical logic gates, quantum logic gates are reversible. That means that quantum gates can leverage two main aspects of quantum mechanics that are entirely out of reach for classical gates: superposition and entanglement. There exists a countless number of quantum logic gates. Some of them were named by various authors. The most common quantum gates operate on vector spaces of one or two qubits, just like the classical logic gates operate on one or two bits. These values determine the probability of measuring a 0 or a 1 when measuring the state-of-the qubit.
Quantum entanglement is a known phenomenon where the properties of two particles are intertwined, despite the great distance from each other. It’s perhaps the strangest quantum property of them all. It’s a sort of quantum marriage between qubits. Entanglement is an extremely useful thing. Measure the value of one qubit and determine the value of a second entangled qubit. However, the more qubits there are, the faster they tend to decohere (see below). NIST (National Institute of Standards and Technologies) claims this is why there are still no quantum computers that can perform beneficial tasks. The system could crash before doing anything worthwhile.
Superposition is a quantum system feature where a particle or electron exists in several separate quantum states at the same time. The electron can be observed in a specific state only when it is measured. That is, it drops out of superposition and adopts one position or the other.
Quantum interference is what allows to bias quantum systems toward the desired state. The idea is to create a pattern of interference where the paths leading to wrong answers interfere destructively and cancel out, but the path leading to the correct answer reinforce each other.
A classical analog for intuition is noise cancellation. Noise cancellation is performed by employing superposition and the principle of interference to reduce the amplitude of unwanted noise by generating a tone of approximately the same frequency and amplitude.
Decoherence destroys superpositions, and it’s a serious problem. So, researchers must protect qubits and isolate them. Decoherence occurs when quantum superposition (an atom in a 0 and 1 state at the same time) collapses by some outside disturbance. It can be a researcher who measures the atom. Decoherence and thermal states are the main errors in QC.
Quantum tunneling is the quantum phenomenon where an atom or a subatomic particle can propagate through a potential barrier and appear on the opposite side of it — something that is impossible for the particle to penetrate. The transmission through the barrier can be finite and depends exponentially on the barrier height and barrier width. Quantum Tunneling is what gives some quantum computers the potential to not only complete tasks faster but to potentially complete tasks a classical computer simply could not do within the confines of classical physics.
Quantum supremacy means that a quantum computer can achieve something that would otherwise be impossible for a classical computer. At the leading edge of the quantum revolution, companies such as IBM, Microsoft, and Google are building computers that aim to do things that classical computers cannot do or could only do in thousands of years. However, the notion of “quantum supremacy” is predicated on assembling a sufficient number of qubits in a single machine.
Who Just Achieved Quantum Supremacy? — The Chinese quantum computing system, called Jiuzhang, completed in 200 seconds what they estimate one of the world’s most powerful supercomputers would have taken 2 billion years to solve.
The problems a quantum computer can tackle more efficiently than a classical computer are called bounded-error Quantum Polynomial-Time (BQP) — a class of decision problems solvable by a quantum computer in polynomial time, with an error probability of at most 1/3 for all instances. A famous example of a BQP problem is factoring.
Unfortunately, most of our problems are NP-complete or NP-hard. These words are a way to say that we don’t know how to solve them and we don’t have the computational resources to solve them. The complex pattern recognition needed for preventing fraud in finance probably falls into the NP-hard category. Problems that can be efficiently solved by quantum computers.
Quantum computing contributes to problems outside the BQP space, while the exact relationship of BQP to NP and PSPACE is not known, as well as approximations, assumptions, and simplifications are needed to transform any problem into other of a lesser order. Using these, we can transform any problem into other of a lesser order. By all means, that comes at the expense of accuracy. Quantum computers allow to approximate less, assume less, and simplify less. Therefore, quantum computers can increase accuracy. The problem of fraud pattern recognition will benefit from this approach.
Recently, Google announced TensorFlow Quantum(TFQ): an open-source library for quantum machine learning, in collaboration with the University of Waterloo, X, and Volkswagen. The aim of TFQ is to provide the necessary tools to control and model natural or artificial quantum systems. TFQ is an example of a suite of tools that combines quantum modeling and machine learning techniques.
1. Convert quantum data to the quantum dataset: Quantum data can be represented as a multi-dimensional array of numbers which is called as quantum tensors. TensorFlow processes these tensors in order to represent create a dataset for further use.
2. Choose quantum neural network models: Based on the knowledge of the quantum data structure, quantum neural network models are selected. The aim is to perform quantum processing in order to extract information hidden in an entangled state.
3. Sample or Average: Measurement of quantum states extracts classical information in the form of samples from the classical distribution. The values are obtained from the quantum state itself. TFQ provides methods for averaging over several runs involving steps (1) and (2).
4. Evaluate a classical neural networks model — Since quantum data is now converted into classical data, deep learning techniques are used to learn the correlation between data.
Quantum Computing — Limitations
At this point, it is also essential to understand the degree of development of quantum computers. We are going through a second quantum revolution, where we are increasingly able to control individual quantum systems to a greater extent than before. That is what drives the development of quantum computing.
However, as exciting and speedy is the second quantum revolution, quantum computing faces monumental challenges. None of the current quantum computers comply with all DiVincenzo’s criteria.
The size of the system determines the scalability, and the difficulty in problem-solving increases exponentially, whereas the datum, is colossal, or the essence of the problem is very complex. The research proves that though quantum computing is a powerful solution to a variety of issues. However, it has limitations and challenges due to dependence on pure physics. However, other technologies advance hardware and software development of quantum computers on which complex algorithms can be created and run.
The frequent challenge that troubles researchers are isolation. Quantum decoherence can arise by heat and light due to constraining such as conditions qubits can lose their quantum properties like entanglement that further leads to a loss in data stored in these qubits. In addition, rotations in quantum computers’ logic gates are prone to error, and they are also crucial to change the state-of-the qubit. The wrong spin can cause an error in the output. The requirement of computers with a greater circuit length and error correction( with redundancy for every qubit) is also crucial for the field of quantum machine learning.
2.11.2 Software Limitations
The developer of algorithms for quantum computers has to be concerned about their physics. While a classical algorithm can be developed along the lines of the Turing machine, to develop an algorithm for quantum computers, the developer has to base it along the lines of physics with no simple formulas that would link it to logic. The critical issue in such a design is always scalability. Design a program to operate on larger data with more processing power. Very little information is available to develop such algorithms for quantum computing. Therefore, most of the development is intuitive. Most known quantum algorithms suffer from a condition of specific simulations that limit their practical applicability and it becomes difficult to develop models that can have a significant impact on machine learning.
Another limitation in quantum computing is that the number of qubits one can have on a quantum circle is limited. Though these limitations apply to quantum computing in general, the augmentation of fields such as machine learning can grab more attention and stimulate the subject with proper gradients and magnitude.
Even those days, experts are still trying to get quantum computers to work well enough, but this task remains challenging, mainly because quantum states are fragile. Decoherence is a critical limitation, where superposition collapses, and the computer loses information. Decoherence results due to the slightest bit of noise in the system. Quantum computers are incredibly complex and expensive machines. As in any emerging technology, the list of limitations goes on and on. With billions of dollars of investment from governments and the world’s biggest companies, the race for quantum computing capabilities is advancing by leaps and bounds.
Critical Milestones for Quantum Computing
Although quantum computing is an immature technology, there are improvements in quantum computing that increase the potential of quantum AI. However, the quantum AI industry needs critical milestones to become a more mature technology. These crucial steps would enable quantum AI for further development.
- Less error-prone and more powerful quantum computing systems
- Widely adopted open-source modeling and training frameworks
- Substantial and skilled developer ecosystem
- Compelling AI applications for which quantum computing outperforms classical computing
According to recent research at the Los Alamos National Laboratory, quantum machine learning can’t be part of investigation processes such as quantum chaos and thermalization (a process that reaches thermal equilibrium through mutual interaction between physical bodies). It places a substantial limit on the learning of any new process linked to it through quantum computing. A study was based on a Hayden-Preskill thought experiment. A thought experiment that investigates the black hole information paradox by hypothesizing on how long it takes to decode information thrown in a black hole from its Hawking radiation.
Quantum Artificial Intelligence — Applications in Financial Crime
Financial Crime is one of the fastest-growing areas of risk management. New technology and regulatory compliance are at the leading edge of this ever-evolving sector, triggering tremendous growth over the past few years as financial institutions seek reliable and trustworthy solutions for fraud, anti-money laundering, and financial markets compliance. Financial services companies harnessing the power of advancement of artificial intelligence and machine learning are setting themselves apart from their competitors by possessing dynamic and sophisticated financial crime solutions.
Financial institutions that can harness quantum computing are likely to see significant benefits. Companies will be able to analyze large structured and unstructured high-dimensional datasets in a centralized and decentralized manner. Sharper insights into the financial crime domain could help banks make better and faster decisions and dramatically improve time-to-value by utilizing the power of quantum computing.
The financial crime sector is beginning to anticipate the colossal transformational possibilities quantum computing could create and priming itself for potential use cases of quantum computers. Many financial institutions and banks are investing in quantum solutions to improve their financial operations. Companies that have invested in quantum computing integration are gaining an advantage in building expertise in financial modeling problems for quantum solutions. As well, financial institutions gain an advantage in hiring already scarce quantum talents.
Head of a research unit at JPMorgan Chase & Co., Marco Pistoia, hopes that quantum computers could potentially boost profits by speeding up asset pricing, digging up better-performing portfolios, and improving existing ML algorithms . A study by a Spanish Bank, BBVA, which came out in July this year, found that quantum computers could boost credit scoring, spot arbitrage opportunities, and accelerate Monte Carlo simulations (The Economist, 2020) . Even the head of quantum research at Goldman Sachs, William Zeng, made a bold claim that quantum computers could “revolutionize” the banking and finance industries (The Economist, 2020) .
Some in the banking industry believe that quantum computing is more science fiction than a fact and that computational power isn’t a key differentiator for the business model. And there is, of course, more to serving clients than computational speed and agility. Still, quantum computing increasingly appears to be a game-changer in tackling complex or intractable problems, particularly in the optimization area. It is only a matter of time before quantum solutions enter the mainstream, which means the window for getting up to speed and gaining competitive advantage will not be open for long.
Financial institutions across the globe are dealing with systemic threats when it comes to financial crime. From fraud and money laundering to know your customer (KYC) and insider trading, these offenses can have a significant impact not only on organizations but also on individuals and economies. They can fuel criminal enterprises and activities, including human trafficking, terrorism, and the drug trade.
Sophisticated criminal groups continue to take advantage of new technologies and fast-changing conditions to carry out increasingly complex and massive illegal schemes, from hiding illicit funds to various forms of first-party and third-party fraud. The net result is that financial institutions face more crime threats than ever, from inside and outside their organizations. As part of their mandated compliance and fraud prevention programs, financial institutions will have a variety of systems and technologies in place. But with complex and ever-changing fraud and financial crime patterns, these tools cannot quickly and effectively stop these incidents without advanced analytics solutions based on AI systems.
The complexity and size of big datasets are growing faster than computing resources and therefore stretch computing capabilities. While today’s computers struggle to solve some problems, quantum computing is likely to solve the same in seconds. And yet, it is predicted that artificial intelligence and machine learning, particularly, can benefit from advances in quantum computing technologies and will continue to do so, even before a complete quantum computing solution is available. Quantum computing algorithms allow us to enhance what’s already possible within the machine learning realm.
In quantum computing, users can physically control parameters like electromagnetic field’s strength, frequency of a laser pulse, or others to solve problems. Accordingly, quantum machine learning algorithms can get trained like artificial neural networks. The main advantage of quantum computers is that they can produce patterns that classical systems thought to have difficulties in production. Therefore, it’s reasonable to assume that quantum computers may outperform classical computers on machine learning tasks. It has led to a new field called quantum machine learning.
One of the ways quantum computing might facilitate a revolution is in its ability to perform data sampling and data processing for exploratory data analysis tasks addressing numerous types of problems such as detecting, preventing and investigating fraud, money laundering, and compliance violations with a holistic view of risk across the organization.
Anisotropic inflation of big data entails the need for different computational architecture approaches to handling enormous amounts of generated data. Not only is it larger in scope, but the categories of the problems that are being tried to solve are very different. Quantum computers are better equipped to solve sequential problems efficiently which is related to data streaming processing and analysis.
Quantum computers will allow quick analysis, data ingestion, and integration from immense centralized and decentralized data sources over topology and boost ETL processes within machine learning development workflows. Quantum computing can handle large datasets at a fast speed and supply data to artificial intelligence-based technologies to analyze it at a more granular level, identifying complex non-repeatable patterns and abnormalities.
Since quantum technologies can enhance machine learning algorithms (quantum-enhanced machine learning) the most common application of quantum computers in the field refers to descriptive and diagnostic analytics within machine learning for data processing that couldn’t be executed through classical computing.
Sampling and Distribution Parametrization
Sampling is a fundamental technique for approximating answers that cannot be computed directly or efficiently due to various constraints such as time, cost, and hardware capacity. Sampling techniques intended to select, manipulate and analyze a representative subset of the dataset to identify patterns and trends in the large dataset being examined. In general, the probability distributions are implicitly defined. It is possible to calculate the probability of any given point easily. But there is no known structure to the distribution. It is not possible to sample precisely according to an arbitrary probability distribution in fewer than classical steps.
Recent theoretical and practical success to demonstrate quantum computations capabilities beyond classical tractability shows that quantum computers can sample from probability distributions that are exponentially difficult to sample using a traditional computer. If these distributions were to coincide with real-world distributions, this would suggest the potential for significant advantage. Recent work on quantum neural networks exhibits this advantage. It seeks to parameterize a distribution through some set of adjustable parameters and quantum kernel methods. These methods use quantum computers to define a feature map that maps classical data into the quantum Hilbert space (infinite-dimensional inner product space having the property that it is complete or closed). The justification for the capability of these methods to exceed classical models often follows similar lines as quantum simulation results. That means it would be effortless to perform intricate sampling processes in situations where it is very hard from traditional computing perspectives.
Quantum computing will lead to giant breakthroughs due to the integration of heterogeneous datasets. Although this may be difficult without human intervention at first, human involvement will help computers learn how to integrate the data in the future. So, if there are different raw data sources with unique schema attached to them and a research team wants to compare them, a computer would have to understand the relationship between the schemas before data comparison. To accomplish this, breakthroughs in the analysis of the semantics of natural language need to happen. And that is one of the biggest challenges in artificial intelligence. However, human-in-the-loop provides data curation, which then trains the system for the future. The promise is that quantum computers will allow quick analysis and integration of colossal datasets that improve and transform machine learning and artificial intelligence capabilities.
Optimization problems are at the heart of machine learning models, decision-making systems-based AI or any other financial crime solution-based advanced analytics. The purpose of optimization is to achieve the best solution under a set of prioritized criteria and constraints. Typical optimization solutions are as follows: include sampling, simulation, minimization of cost functions, fitting machine learning algorithms, training models, regularization, HPO, and much more.
Quantum computers have unique capabilities that lend themselves to possibly solving these complex types of optimization problems more efficiently: using the quantum property of superposition to represent all possible solutions while quantum property of interference enables the identification of low-cost, high-value solutions. Classical methods for these problems tend to have either exponentially growing to compute times or suboptimal performance. Quantum optimization algorithms such as the quantum approximate optimization algorithms (QAOA) hold the promise of finding the answers that improve on the suboptimal solutions without incurring the cost of exponentially longer computing times. It could include finding better optima in a training landscape, performing non-convex optimization, escaping or resolving saddle points (where the function has a local maximum in one direction but a local minimum in another), or finding optima with fewer queries.
D-Wave (a Canadian quantum computing company), is extensively using quantum annealing to solve optimization and sampling problems. Recently they launched an open-source plugin that allows developers to easily map quadratic optimization inputs in IBM’s Qiskit format onto D-Wave’s quadratic unconstrained binary optimization (QUBO) format and solve the same input on any quantum system supported in Qiskit. This allows users to have a realistic assessment of the benefits of quantum annealing on their applications.
Quantum annealers are a sort of universal quantum computers version that specializes in finding super-local minima and closer approximations to a global minimum than a classical computer. Quantum annealers work by having a series of magnets attached to a grid. The magnets affect one another and by inverting into a coordinated orientation save energy for the system by minimizing its usage. In the traditional systems, the magnets are confined into low-energy settings before being able to find lower minima. However, attaining quantum properties such as tunneling, they can omit those large energy cost settings that allow for functions to more easily descend from a local minimum into either a global minimum or to the closest local minima near the global minima.
When it comes to cost functions, this can mean the difference between a gradient descent function being stuck in a suboptimal setting to one where it is optimal or near-optimal, especially on complex non-convex error surfaces where saddle points might be base.
This approach provides a solid solution to complex machine learning optimization problems in case there are soft requirements. That means a situation where enormous traditional computing power is required. By exploiting quantum tunneling, there is a quick convergence to minimization of the error function in optimization applications such as fraud loss analysis in the financial crime domain.
Monte Carle Simulation
Monte Carlo Method is a mathematical technique is used to estimate the possible outcomes of an uncertain event. Monte Carlo simulation performs random sampling to estimate numerical quantities that are hard to compute deterministically. It calculates results repeatedly, using a different set of random values from the probability functions. Monte Carlo simulations are used to model the probability of different outcomes in a process that cannot easily be predicted due to the intervention of random variables.
It is a technique used to understand the impact of risk and uncertainty in prediction and forecasting models in financial crime solutions. However, to obtain the most efficient result with a small associated error the required number of simulations is gigantic. It is common for these computations to take up to 24 hours.
Consequently, obtaining a quantum speedup in the Monte Carlo Method can be very rewarding. It can be achieved using a quantum amplitude amplification estimation algorithm (amplitude amplification) that can sample a probabilistic distribution quadratically faster than the classical methods. The Quantum amplitude estimation algorithm employs the maximum likelihood estimation (MLE) based on measurement data produced from quantum circuits with a different number of amplitude amplification operations.
Risk analysis and pricing of financial derivatives are some of the problems which can be tackled using this method. Determining the risk of the complete portfolio of a desk often requires overnight calculations of the prices. Through quantum computing, overnight calculations could be reduced to much shorter time scales (such as minutes). It allows the institution to react faster to changing market conditions and to profit from time-to-value opportunities.
IBM has already published a paper that showcases a quadratic accelerated version of Monte Carlo simulations. Cambridge Quantum Computing (CQC) recently made a breakthrough in quantum monte Carlo integration using quantum amplitude estimation. JPMorgan Chase and Barclays have been using IBM’s quantum-computing software to test Monte Carlo simulations for portfolio optimization since 2017. Goldman Sachs was able to speed up derivatives pricing by a thousand times for a million potential paths compared to the Monte Carlo method. BMO Financial Group and Scotiabank collaborated with Xanadu to benchmark a quantum Monte Carlo algorithm for a variety of trading products.
Building and training machine learning models impose substantial challenges. A lot of computing resources are needed to train robust and generalized models. Simple datasets can be trained using only the CPUs and should take only a couple of minutes. More complex deep learning models, on the other hand, require anywhere from 2 to 32 GPUs and u to several days to train, especially when dealing with typical financial crime datasets that are high-dimensional, sparse, and extremely imbalanced due to the minority class of fraudulent transactions.
The development of quantum algorithms for quantum generalizations of classical learning models can provide feasible speed-ups and other improvements in the deep learning training process. Quantum computing can be used for the rapid training of machine learning models and to create optimized algorithms. The contribution of quantum computing to classical machine learning can be achieved by quickly presenting the optimal solution set of the weights of artificial neural networks. An optimized and stable AI provided by quantum computing can complete years of analysis in a short time and lead to advances in technology.
Quantum computing is expected to search very largely, unsorted datasets to identify patterns or anomalies extremely quickly. It might be possible for the quantum computers to access all items in the database simultaneously to identify these similarities in seconds. While this is theoretically possible today, it only happens with parallel computing looking at every record sequentially, so it takes an incredible amount of time, and depending on the size of the dataset, these kinds of tasks might never be accomplished.
The creation of quantum computing architectures is full of challenges that are hard to solve by traditional discrete computation technologies and are better suited for AI models. One of those problems is known as quantum control optimization and focuses on the design of quantum controls to translate each quantum algorithm into a set of analog-control signals that accurately steer the quantum computer around the Hilbert space (vector space equipped with an inner product operation, which allows lengths and angles to be defined). The precise choice of these controls ultimately governs the fidelity and speed of each quantum operation. Until now, quantum computing has been lacking a universal control framework that facilitates optimization over major experimental non-idealities under systematic constraints which have constrained the creation of quantum architectures.
In a paper titled “Universal Quantum Control through Deep Reinforcement Learning” Google proposed a new deep reinforcement learning framework that simultaneously optimizes the speed and trustworthiness of quantum computation against both leakage and stochastic control errors. Control robustness improved by adding control noise into training environments for reinforcement learning agents trained with trusted-region-policy-optimization. The agent control solutions demonstrate a 2-order-of-magnitude reduction in average-gate-error over baseline stochastic gradient descent solutions and up to a 1-order-of-magnitude reduction in gate time from optimal gate synthesis counterparts.
Quantum cloud computing is a quantum computer that can be accessed in a cloud environment through a network. Cloud-based quantum computing provides direct access to emulators, simulators, and quantum processors. One of the biggest advantages of availing quantum services in the cloud is that it allows access to quantum physical-powered computers via the web. In 2016, IBM connected a small quantum computer to the cloud, and it allows for simple programs to be built and executed on the cloud. In early 2017, researchers from Rigetti Computing demonstrated the first programmable cloud access using the pyQuil Python library. The Quantum Orchestration Platform offers a unified framework for controlling quantum processors and running heavy classical processing. An advanced stream-processing framework and built-in AI engines allow running and optimizing hybrid quantum-classical algorithms from day one.
Quantum computing in the cloud has the potential to disrupt industries in a similar way as other emerging technologies, such as AI and machine learning. The cloud services today are aimed at preparing the industry for the soon-to-arrive day when quantum computers will begin being useful. Quantum computing could even supplement general compute and AI services cloud providers currently offer. In that scenario, the cloud would integrate with classical computing cloud resources in a co-processing environment. In fact, there is a lot of potential for quantum computing and artificial intelligence symbiosis leveraging the elastic nature of the cloud and the powerful, problem-specific solving capabilities of quantum computing. The conceivable prospect is to perceive the two working together harmoniously to solve challenging and complex problems. Both have strengths and for quantum computing to function as part of the solution. Over time, both computing formats will continue to advance, but the ability to accelerate workloads on traditional GPUs and ASICs while also leveraging the power of quantum computing is a recipe for faster, more robust results, which is what the market should be eager to see as quantum computing becomes more widely accessible.
One of the most eminent applications of cloud-based quantum computing is leveraging elements of de-centralized AI such as federated and distributed machine learning paradigms. Both model-centric and data-centric approaches might gain more speedups and robustness while training middle-large dataset batches in an offline or online learning manner by incorporating collective intelligence over the cloud.
Quantum Neural Networks
Quantum neural networks are a subclass of variational quantum algorithms, comprising quantum circuits that contain parameterized quantum gate operations. Information is first encoded into a quantum state via a feature map that adapted enhanced performance of the quantum model and is typically neither optimized nor trained. Once data is encoded into a quantum state, a variational model containing parameterized quantum gates is applied and optimized for a particular task. This happens through loss function minimization, where the output of a quantum model can be extracted from a classical post-processing function.
Quantum neural networks (QNN) is an emerging deep learning paradigm that promotes the creation of artificial neural networks that can run on quantum computing architectures. In a paper “Classification with Quantum Neural Networks on Near Term Processors”, Google proposed a quantum neural network model focused on classification tasks that represent labeled data. The proposed architecture differs from classical deep neural networks. Instead of hidden layers, a quantum neural network architecture is structured by quantum gates on qubits. Google trained their quantum neural networks in the famous MNIST dataset and the results were very impressive.
Training artificial neural networks is a notoriously tricky task, however, quantum neural networks exhibit resilience and are trained faster than classical artificial neural networks models due to their favorable optimization landscapes, captured by evenly spread of Fisher information (how much information about an unknown parameter we can get from a sample) spectrum.
Financial crime solutions based on quantum neural networks might benefit from a powerful, generalized model that can capture perplexing fraud patterns within volume data and can be efficiently trained in minutes. By entailing variational quantum algorithms, quantum neural networks architecture drive towards the realization of fraud solutions based on data consortium over the cloud by incorporating distributed deep neural networks.
Explainable Artificial Intelligence
Model governance, risk, and compliance are crucial for financial institutions in assuring the reliability of operations and achievement of business results. There is increasing pressure on financial institutions and businesses to have proper risk management frameworks in place to detect, prevent and investigate financial crime. Hence, financial crime solutions should meet the requirements of regulation and provide explainable, transparent, and trustworthy analytical solutions.
One unique characteristic of the quantum logic gates is that they are reversible, which means that, unlike classical logic gates, they come with an undo button. In practical terms, this means they never lose information up to the point of measurement, when qubits revert to behaving like classical bits. It can be useful in the area of explainability and interpretability of machine learning models. A feature importance set that determines an algorithm’s prediction of a fraudulent transaction is easily accessible.
Using innovative technology to protect institutions and safeguard both consumers’ and investors’ assets, NICE Actimize identifies financial crime, preventing fraud and providing regulatory compliance. It provides real-time, cross-channel fraud prevention, anti-money laundering detection, and trading surveillance solutions that address payment fraud, cybercrime, sanctions monitoring, market abuse, customer due diligence, and insider trading. AI-based systems and advanced analytics solutions find abnormal behavior earlier and faster, eliminating financial losses from theft to fraud, regulatory penalties to sanctions. As a result, organizations reduce losses, increase investigator efficiency, and improve regulatory compliance and oversight.
NICE Actimize sees direct applications of quantum artificial intelligence where quantum computing can take advantage of speed and specificity to help overcome very complex problems within artificial intelligence and machine learning. In fact, quick wins for quantum computing are most likely in subdomains of artificial intelligence such as machine learning has already improved traditional classification and regression solutions applying predictive analytics for prediction and forecasting.
NICE Actimize believes that over the next few years, quantum computing is likely to supercharge these activities. To be sure, this will be a long road, and most banks are taking their first steps. However, there are three actions for banks to consider now as they begin the journey:
- Build research partnerships with QAI developers such as Amazon, D-Wave, IBM, Google, Microsoft, Rigetti, and Xanadu. These companies have the hardware and expertise to help organizations develop their capabilities.
- Create a small team focused on quantum computing. In financial crime solutions based on AI and advanced analytics, these kinds of partnerships are not one-way streets, and quantum providers are also keen to learn from financial industry players about their algorithms and requirements. Make certain to contribute to collaboration.
- Explore potential investments or joint ventures. Joint ventures are presumably to be prevalent considering the challengers of building quantum systems. Classical solutions fused with quantum technologies may form cloud offering and merging complex optimizations for data analysis and simulation. It might be obtained throughout a vendor platform that can smooth the adoption process and assist in running initial POCs.
- Early adopters are conceivably to have superiority. Regardless, in all cases, a practical first step is to rewrite internal algorithms in quantum language, which will pave the groundwork for substantive funding.
The arrival of quantum computing is potentially game-changing. Financial institutions are only just starting to get access to the necessary hardware and to develop the quantum algorithms they will need. Still, a rising number of initiatives suggest that a Rubicon is on the horizon. For banks yet to engage, and particularly those that rely on computing power and AI-based system to generate a competitive edge, the time to act is now.
In the coming decades, quantum computation will play an essential role in technology, science, and business advancement. With Google claiming quantum supremacy and exponential interest in this subject, we aren’t far from seeing many intractable issues challenging to tackle with a classical computer solved by quantum computing. Furthermore, there is a clear indication that machine learning and quantum computing will play complementary roles in strengthening each other’s areas.
Some in the banking industry believe that quantum computing is more science fiction than fact, and that computational power isn’t a key differentiator for the business model. And there is, of course, more to serving clients than computational speed and agility. Still, quantum computing increasingly appears to be a game-changer in tackling complex or intractable problems, particularly in the optimization area. It is only a matter of time before quantum solutions enter the mainstream, which means the window for getting up to speed and gaining competitive advantage will not be open for long.
- Alexander Streltsov et. al., (2017) Colloquium: Quantum coherence as a resource. Rev. Mod. Phys., 89:041003
2. Amira Abbas et.al., (2020). The Power of Quantum Neural Networks
3. Bernstein, E., & Vazirani, U. (1993, June). Quantum complexity theory. ACM Digital Library.
4. C. H. Bennett, et.al.,(1997). Strengths and weaknesses of quantum computing. SIAM Journal on Computing, 26(5):1510–1523
5. D. M. Bacon. (2001) Decoherence, Control, & Symmetry in Quantum Computers. University of California at Berkeley
6. D.V. Fastovets et. al., (2019). Machine Learning Methods in Quantum Computing Theory
7. E. G. Rieffel and W. H. Polak (2011) Quantum Computing: A Gentle Introduction. The MIT Press, 1 edition
8. E. Tang. (2018) A quantum-inspired classical algorithm for recommendation systems. Electronic Colloquium on Computational Complexity (ECCC), 25:128
9. Edward Farhi et. al., (2018). Classification with Quantum Neural Networks on Near Term Processors
10. Eleanor G., Quantum Supremacy Using a Programmable Superconducting Processor, NASA Ames Research Center
11. Faye, J. (2002, May 3). Copenhagen Interpretation of Quantum Mechanics. Stanford Encyclopedia of Philosophy
12. Feynman, R. P. (1986). Quantum mechanical computers. Foundations of Physics, 16(6), 507–531.
13. Gilles Brassard et.al., (2000). Quantum Amplitude Amplification and Estimation
14. Hsin-Yuan Huang et.al., (2021, Google research). Power of data in quantum machine learning.
15. Ivan H. Deutsch. Harnessing the power of the second quantum revolution. PRX Quantum.
16. John Preskill, (2020) Institute for Quantum Information and Matter, California Institute of Technology Quantum Computing 40 years Later.
17. Kopczyk, D. (2018, April 25). Quantum machine learning for data scientists
18. Murphy Yuezhen Niu et. al., (2018).Universal Quantum Control through Deep Reinforcement Learning
19. S. Simon (2010) Quantum computing…with a twist. Physics World
20. Vadlamudi, S. (2019). How Artificial Intelligence Improves Agricultural Productivity and Sustainability: A Global Thematic Analysis. Asia Pacific Journal of Energy and Environment,6(2), 91–100.
21. Weiwen Jiang et. al., (2020). When Machine Learning Meets Quantum Computers: A Case Study