Top Applications Of Quantum Computing for Machine Learning

Machine Learning has two roles within quantum computing. On the receiving side, quantum computers use classical machine learning to optimize hardware operations, control systems, and user interfaces. But, on the giving side, quantum computers perform machine learning tasks and solve traditional machine learning problems. Focusing on the latter, some of the most commonly cited quantum machine learning applications are:

• Enhance algorithms designed to solve optimization problems, which in turn have a broad range of potential applications across many industries
• Accelerate classification tasks for large datasets, particularly related to image and speech recognition, with quantum support vector machines (QSVM)
• Enable exploration of high-dimensional datasets, such as customer segmentation and anomaly detection, using algorithms such as K-means clustering
• Reduce dimensionality, using principal component analysis, to improve feature selection and thus improve the visualization and analysis of data
• Generate more realistic data for Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) when the data either doesn’t exist or needs to be augmented
• Discover complex patterns and correlations using Boltzmann machines for unsupervised learning, which has utility in recommendation systems and natural language processing
• Improve neural network training by leveraging interference, potentially saving time by reducing the number of iterations required to find the optimal parameters
• Solve near-term quantum chemistry problems, particularly applied to discovering new drugs and developing new materials, but also applied to a broad range of other problems
• Analyze data for features and anomalies, the latter of which has important implications for network traffic analysis and fraud detection for financial services
• Optimize policies for autonomous systems, including self-driving cars and unmanned aerial vehicles (UAV), using reinforcement learning

Although quantum computing is still in its infancy, cloud-based quantum machine learning solvers are already available. Some are proprietary and commercialized, but there are also open-source quantum machine learning services that are freely available.

What is Quantum Machine Learning?

Quantum Machine Learning (QML) is an interdisciplinary field that applies quantum computing to traditional machine learning tasks. But, classical machine learning already does that, so quantum machine learning algorithms must propose advantages over their classical counterparts:

• The basic unit of information, the qubit, is not a 0 or a 1, but a complex number; depending on several factors, a complex number is minimally several bytes, each of which is eight 0s and 1s.
• Entanglement exponentially increases the number of complex numbers required to describe a system, which means exponential compression of classical data
• Quantum computation is not sequential like classical computation; it is inherently parallel, which allows multiple solutions to effectively be discovered simultaneously
• Theoretical research supports the possibility of achieving significant computational speedups over known classical approaches
• Data can be mapped not only from classical sources but also from quantum sources such as the proposed Quantum Random Access Memory (QRAM)
• Interference can be leveraged to not only increase the probabilities of receiving correct solutions, but also to potentially provide speedups over classical algorithms

A paper by D.P. García, et al, titled “Systematic Literature Review: Quantum Machine Learning and its applications” reviewed 52 articles, identifying and describing 18 hybrid classical quantum machine learning algorithms as a result. These algorithms include Boltzmann machines, parameterized circuits, autoencoders, reservoir computing, support vector machines (SVM), and an assortment of neural networks. This diversity of algorithms highlights the potential breadth of real-world problems that might be solvable with quantum machine learning.

{{Deloitte-Webinar}}

Quantum Advantage in Machine Learning

The term “quantum advantage” typically refers to speed, but the term is not strictly defined. When a claim is made that a computation can be performed in 200 seconds that would take a supercomputer thousands of years, that’s a claim of quantum advantage. But, that suggests that there’s only one classification of quantum advantage when, in fact, there are multiple potential advantages:

• Speed. The “holy grail” of quantum computing is to solve problems that consume considerable time and significant classical resources.
• Precision. Many hard problems can only be approximated classically, whereas quantum algorithms can offer precise solutions.
• Complexity. Quantum algorithms may solve problems in fewer timesteps than classical algorithms, indicating greater efficiency.
• Compression. Large datasets that require prohibitive amounts of classical memory may be mapped to a relatively small number of qubits.
• Dimensionality. As an extension of data compression, quantum computers are a natural fit for finding patterns and relationships in high-dimensional data. 
• Sampling. Quantum computers can naturally sample probability distributions for a wide range of applications, including generative algorithms.
• Interference. Constructive and destructive interference can be leveraged to increase the probability of finding correct solutions and decrease the probability of incorrect solutions.

It’s important to note that quantum algorithms are not guaranteed to be advantageous in any way. Nor can they be expected to realize all of the advantages above. However, the potential to realize these advantages exists, and some quantum machine learning solutions might prove to be exceptionally advantageous.

‍

{{Newsletter-signup}}

‍

Quantum Machine Learning Applications

Quantum machine learning (QML) use cases overlap two other major classifications of quantum computing applications: quantum simulation and quantum optimization. Near-term solutions for both incorporate hybrid classical-quantum algorithms that utilize classical neural networks. And anywhere you find a classical neural network, is a potential application of quantum machine learning, as well:

• Molecular simulation, the original application of quantum computing, with use cases in drug discovery, material development, and climate modeling
• Optimization problems, with particular interest in financial applications, but with broad applicability across many industries, especially those with supply chains
• Natural language processing, including the quick and accurate interpretation, translation, and generation of spoken languages
• Imaging, including the classification and identification of objects in pictures and videos, and with applications ranging from health care to surveillance to autonomous vehicles
• Machine Learning model training, optimizing how artificial neural networks train on classical data to solve the problems above 

‍

Specific examples of some of these use cases, broken down by industry, include:

• Energy sector, including energy grid optimization, energy demand prediction, energy storage, energy storage distribution, and renewable energy integration
• Manufacturing, including production scheduling, shift scheduling, resource allocation, and defect detection, but also the optimization of individual processes
• Retail sales, including demand forecasting, inventory management, delivery route optimization, recommendation systems, loss prevention, and store layout optimization
• Government services, including traffic optimization, public transportation routing, policy modeling and simulation, and public health modeling
• Finance and insurance, including financial portfolio optimization, financial risk management, fraud detection, and trading strategy optimization
• Aerospace and space, including trajectory optimization and flight data analysis, both terrestrial and extraterrestrial, and communication optimization
• Environment and weather, including weather forecasting, disaster prediction, environmental modeling, and natural disaster response management
• Health care services, including medical image analysis, diagnosis assistance, treatment plan optimization, early detection screening, and clinical trial design
• Customer service and social media, including personal recommendations, sentiment analysis, support ticket routing, omnichannel experience optimization, and chatbots

Although no references are provided, an Analytics Insight article titled “What is Quantum Machine Learning? Applications of Quantum Machine Learning” proposes some novel quantum machine learning solutions. Some overlap what is commonly referred to as “quantum simulation,” but they also list human anatomy, space exploration, and cybersecurity.

Most cloud based quantum machine learning options are impractically small due to low qubit counts, but Aquila allows exploration of Quantum Machine Learning with 256 qubits. Neutral atoms are a natural fit for solving Quantum Optimization problems, and the two fields are intertwined, so the research prospects are very promising.

Adapting classical machine learning to the quantum domain

Many classical machine learning algorithms do not naturally map to the quantum domain. This may be for a number of reasons:

• The algorithm requires a large number of shots, which may be significantly slower on a quantum computer
• The algorithm seeks to optimize the parameters using methods such as gradient descent, but quantum measurements are noisy
• Quantum algorithms do not yet support millions of parameters such as classical ones.

Because of this, new QML algorithms are being developed to take advantage of the unique properties of quantum while sidestepping quantum’s limitations.

‍