What is Quantum Machine Learning?

Quantum machine learning is the field of research that combines machine learning and quantum computing technologies. Quantum computing utilizes specialized processors containing quantum particles to solve complex problems faster than classical computers. As quantum particles operate at the sub-atomic level, they require a new set of principles and mathematical equations called quantum algorithms. Quantum machine learning involves both—using existing quantum algorithms for solving machine learning problems and tweaking machine learning algorithms for quantum hardware. Quantum machine learning research aims to find new solutions to existing prediction and classification problems currently solved by statistical techniques over neural networks.

Traditional machine learning trains computer software to analyze data and make accurate predictions. To do that, data scientists develop machine learning models, which are software functions capable of training from datasets. Some of these models require substantial processing power to train, operate, and scale. For example, deep learning algorithms run on supercomputers and process massive datasets. Quantum machine learning expands the processing capabilities of traditional machine learning beyond a supercomputer's limits.

Advanced hardware

A quantum computer stores data as qubits on specialized hardware. Qubits are similar to binary bits that traditional computers use to store data. You can imagine qubits as storing the various sides of a coin as it's flipped into the air, while binary bits capture the resting coins on the ground. Qubits allow superposition and entanglement of stored information.

• Superposition represents data in more than one state.
• Entanglement is the ability of a quantum particle to interact and exchange with other particles.

Unlike traditional computing, which stores data in either 1 or 0, a quantum device can represent data in a state of uncertainty. These properties, along with quantum computation, allow quantum systems to perform complex machine-learning tasks that supercomputers struggle with.

Data processing speed

Quantum computers apply quantum theory to processing and storing information. Quantum computation is a quantum computer's ability to process information in parallel. Unlike sequential processing in classical computers, quantum computation and quantum speedup significantly reduce data processing time. For example, quantum computers could take minutes to solve problems that classical computers took thousands of years to solve.

Quantum machine learning is an evolving field that will shape machine learning development in the near future. We share several benefits of quantum machine learning techniques below.

Improves machine learning prediction

Quantum computing allows deep learning models to analyze massive datasets faster and more accurately. For example, data scientists can develop quantum support vector machines (QSVM) to classify images. QSVM uses quantum computation to process input data in parallel to predict an outcome quickly.

Analyzes high dimensional data

High-dimensional data are datasets that contain more features than observations. For example, a medical dataset may contain more health parameters (features) than the number of patients (observations). Quantum machine learning efficiently extracts features from high-dimensional datasets because of its exponentially superior computational power. Theoretically, the quantum neural network can improve generalization speed by storing data in quantum random access memory (QRAM). This allows organizations to apply dimensional reduction techniques, such as quantum principal component analysis, more efficiently.

Enable quantum reinforcement learning

Reinforcement learning is a technique that optimizes a model's behavior with rewards or penalties through self-learning. With quantum reinforcement learning, autonomous systems can learn more efficiently from their immediate environment. For example, self-driving cars powered by quantum neural networks can improve their navigational ability in fewer iterations.

Develops advanced machine learning algorithms

Current machine learning models are limited by the computer they run on. With quantum computing, data scientists can design powerful machine learning models with a significant quantum advantage over classical algorithms. This introduces possibilities for advanced machine learning applications in drug discovery, computer vision, and other industries.

Analyses complex data patterns

Quantum machine learning systems detect and correlate complex patterns more easily than classical neural networks. It is helpful in innovating technologies like natural language processing, image recognition, and recommendation systems.

Quantum mechanics is a science that studies the smallest form of particles and their characteristics. It explores interactions and behaviors of sub-atomic particles, including how energy transfer happens through wave-like properties. Quantum machine learning integrates classical machine learning with quantum mechanics. It involves concepts not previously found in conventional machine learning methods.

Quantum computing concepts

As quantum computing uses specialized hardware, new hardware-related concepts emerge in quantum machine learning.

Quantum circuits

Quantum circuits are electronic components designed to operate with quantum physics. They consist of several quantum gates, similar to logical gates in conventional electronic components. A quantum gate accepts an input and transforms it into an output.

Quantum state

Quantum state refers to an object’s possible states in a quantum environment. Rather than existing in a fixed state, the quantum state is ever-changing. Think of it like a wave that increases or decreases in phase and magnitude instead of fixed values.

Quantum data

Quantum data are information stored in a quantum system. Quantum computers generate and store quantum data in qubits. Because quantum data can be in several states simultaneously, processing quantum data requires immense computing power.

Quantum machine learning concepts

The below concepts relate to the application of machine learning in quantum computing.

Quantum-classical model

A quantum-classical model is an advanced machine learning algorithm that runs classical machine learning methods on a quantum system. It leverages quantum supremacy to analyze data accurately and quickly.

Quantum kernels

Kernels are a collection of machine learning functions that implement analysis and classification tasks. They help separate non-linear data into two more regions. Quantum kernels use a quantum feature map to represent high-dimensional data. These data might appear noisy to classical machine learning algorithms, but quantum machine learning can extract, analyze, and classify them with high accuracy.

Quantum transfer learning

Transfer learning is a machine learning technique that allows a model to apply its knowledge to a related task. For example, a machine learning model trained with images of cats can reuse most of its knowledge to identify dogs instead of learning from scratch. Quantum transfer learning is similar, except that it replaces classical neural networks with quantum circuits.

There are two main approaches in quantum machine learning research.

Using quantum algorithms as neural networks

The quantum computing community has developed quantum algorithms, which are practical methods for utilizing qubits for computation. These algorithms are based on the same principles that allow using neural networks in traditional machine learning. In the same way that the weights of a neural network are iteratively adjusted based on the error in the training set, the quantum algorithms use classical optimization routines to iteratively adjust the parameters of a quantum circuit, like the rotation angle in a qubit gate.

For example, the variational quantum algorithm undergoes training before it can make accurate predictions. During training, data scientists adjust the quantum circuit's parameter based on the model's gradient function to achieve optimal prediction.

Scientists are researching the best ways to optimize quantum hyperparameters for machine learning. They use approaches like classical optimization algorithms and reinforcement learning to configure parameters such as the number of cubits and quantum circuit depths.

Running ML models on quantum hardware

Another approach that scientists are exploring is designing, training, and running classical models on quantum computers. While commercially viable quantum computers are yet to be available, data scientists have studied the potential of running existing neural network software on quantum devices. Early studies show that classical neural networks can be exponentially faster when processing data stored in qubits. Moreover, quantum hardware allows neural networks to predict an outcome in much fewer steps than a conventional computer. For example, the quantum distance estimation algorithm allows a classifier model to complete machine learning tasks faster in a quantum computer. Also, neural networks are more expressive when running on quantum computers. Such machine learning models can interpret and correlate complex data relationships more efficiently, extending their use cases into applications like fraud prevention and recommendation systems.

Many theories on quantum machine learning have remained unproven. In recent years, scientists have been actively contributing ideas, research, and experiments to prove that quantum machine learning does have a substantial advantage over classical models. Proving quantum advantage is challenging because developing quantum computers requires futuristic technologies we don't possess yet. However, scientists have demonstrated promising results with quantum machine learning algorithms and circuits.

Quantum technologies at AWS offer services to help you explore how quantum computing can impact your business. For example, Amazon Braket is a fully managed quantum computing service designed to help speed up scientific research and software development for quantum computing. With Amazon Braket, you can:

• Learn how to program quantum computers and explore potential applications.
• Design your quantum algorithms from scratch or choose from pre-built algorithms.
• Test your implementation on fully managed quantum circuit simulators.

When ready, you can run your algorithm using our hardware providers’ different quantum computers or quantum processing units (QPUs). With Amazon Quantum Solutions Lab, you can work with leading experts in quantum computing, machine learning, optimization, and high-performance computing.

Get started with quantum machine learning on AWS by creating a free account today.