The power of quantum machine learning
It is said that quantum computing will soon revolutionize many areas of technology – amongst others machine learning. But how exactly are quantum computing and machine learning combined? And what impact can this combination have? Let’s do a deep dive into the field of quantum machine learning and find out why you should care about it now.
by Philip Zupancic
How can we make radiation therapy less damaging to surrounding healthy tissue and body parts?
How can we warn people from dangerous storms well in advance?
And how do we successfully protect autonomous vehicles from hacker attacks?
These are all topics that can – in terms of methodology – be tackled with machine learning (ML), more precisely, ML powered simulations, predictions and computer vision. However, the challenges as they are put here are so complex that to solve them, we need powerful computing resources that can support our machine learning methods. In this context, quantum machine learning (QML) is a promising approach.
But what is quantum machine learning exactly? How does it work? And why should you care about it now? That’s what you’ll find out in this article.
What are quantum computers? And what is quantum machine learning?
Before we can talk about quantum machine learning, we first need to establish what quantum computers are. One of the most important differences between quantum computers and “classical computers”, as we will call them, is that quantum computers work with qubits instead of bits (see Figure 1). While a bit can only represent 2 states (either 0 or 1), a qubit can represent 0, 1, or any weighted combination in between (this phenomenon is called superposition). This way, a quantum computer can support multiple states at the same time and, through this massive parallelization, perform complex calculations much more quickly than a classical computer.
Figure 1: A classical computer works with bits, while a quantum computer works with qubits that can represent in-between states of 0 and 1.
Having said this, the term quantum machine learning simply stands for the combination of quantum computing and machine learning. This can go in two directions:
- Quantum computers support machine learning processes: for example, they can speed up the time it takes to train or evaluate a machine learning model.
- Machine learning methods further quantum computing: machine learning methods can be used to develop new quantum algorithms or to uncover codes with which we can correct errors in quantum calculations.
In this article, we will focus on the first point, also called “quantum-enhanced machine learning”. So, let’s look at how exactly quantum computing can make our machine learning models more powerful.
How does quantum machine learning work?
Quantum computers can – among other things – speed up processes that would take a classical computer a much longer time, e.g., train a machine learning model to solve a highly complex task.
To do this, a quantum computer is often integrated into the machine learning process as a co-processor: parts of the machine learning algorithm are run on a traditional computer; however, computationally difficult subroutines, which the classical computer would struggle with, are outsourced to a quantum device which can execute more complex calculations faster. Thus, quantum machine learning solutions are mostly hybrids in which classical (CPUs) and quantum processing units (QPUs) work together.
An example for this is the variational quantum CNN (convolutional neural network) shown in Figure 2. This network has the task of classifying images – in this case: of deciding whether an image shows a cat or not. Convolution and classification are performed on quantum hardware by entangling the input qubits and measuring the results. The classical computer compares the input labels to the output labels and tunes the configuration of the quantum circuit to optimize the performance (this is the ”variational” part), thus training the network.
Figure 2: A variational quantum CNN, implemented as operations on qubits on a quantum computer (check out this article for more information).
The interesting result: By combining the strengths of quantum computing and classical computing in this particular use case, the variational quantum CNN needs to see fewer images to work well on unfamiliar data. Or in other words: the model generalizes better than its classical counterpart.
This is just one example for quantum machine learning from a variety of envisioned and tested architectures. Numerous studies demonstrate that quantum neural networks need significantly fewer parameters for the same performance compared to classical ones (95% less in this article), which can lead to faster training and more robust convergence. Other approaches aim to exponentially accelerate the algebra in conventional machine learning to speed up both training and model execution. We cannot yet say with certainty under which conditions quantum machine learning outperforms its classical counter parts “in the real world”, but the results published so far are promising.
After having found out what characterizes quantum computing and how it can be combined with machine learning, let us now return to the example challenges from the beginning of the article to find out how exactly quantum machine learning can solve them.
GET READY FOR THE QUANTUM IMPACT
According to the market research firm Tractica, spending on quantum computing will grow from $260 million in 2020 to $9.1 billion in 2030. Businesses can start their own quantum journey with 4 simple steps.
Read more in this report.
How can quantum machine learning be applied?
At the beginning of this article, we established that quantum machine learning can help with challenges like making radiation therapy less harmful for healthy body parts, recognizing the advance of dangerous storms well in advance and protecting autonomous vehicles from hackers. Let’s look at how these real-life challenges can be solved in more detail:
- QML improves radiation therapy and medical diagnostics.
Radiation therapy usually damages the healthy tissue and body parts surrounding the collection of cancer cells. In order to minimize this damage, we need to define an optimal radiation plan. This requires a highly personalized treatment and many iterations of machine learning simulations. There are indications that the increased resistance to noise and better pattern recognition ability of quantum-enhanced algorithms can assist in this process. Similarly, medical diagnostics are apt to benefit from feature selection, bio marker discovery and risk level assessments through quantum machine learning.
- QML helps forecast weather developments.
To predict weather developments such as storms in advance, we use simulations that are produced by so-called generative machine learning models. These models depend on countless parameters such as the state of the atmosphere, land and ocean. This makes it difficult to predict weather phenomena in the far-away future. The structure of quantum networks seems well equipped to handle this complexity: In a recent study, researchers used a hybrid quantum approach and synthetic data produced by a supervised quantum machine learning model to predict storms. According to the study, the hybrid model performed as well as a classical baseline model. Moreover, replacing one layer of the classical neural network with a quantum convolutional layer already improved the model’s predictions.
- QML make autonomous vehicles safer.
Machine learning has shown great potential for advancing self-driving vehicles, but can struggle in this area due to the complexity and limited processing power. There is an increasing interest in employing quantum machine learning for a wide range of applications including object and road sign detection, decision making, and protection against adversarial attacks. Autonomous vehicles are also feared to become popular targets for hackers. Quantum and post-quantum cryptography can ensure the security of vehicular systems.
Why should you care about quantum machine learning now?
Quantum computing – and with that also quantum machine learning – is only just beginning to unfold its potential. Currently, quantum computers exist with a few hundred qubits, which is enough for proof-of-concept studies, but not to tackle relevant real-world problems that current high-performance computers are struggling with. However, the road ahead in terms of hardware development is well established, and companies are quickly developing ever more powerful quantum computers. Indeed, availability and usage thereof is only a matter of time.
Consequently, companies as well as governments are investing heavily into quantum computing. This leads to a fast-growing ecosystem of hardware as well as software providers. Organizations that want to make use of this powerful new technology are therefore well advised to inform themselves now about what benefit this can bring their business and how they best prepare. To get started, I recommend reading this article for some business perspectives. If you are interested in seeing code examples and getting started with hands-on quantum machine learning, check out Qiskit, Pennylane or Tensorflow Quantum. Happy coding!
SIMULATING ICE FORMATION – WITH QUANTUM ACCURACY
Materials scientist Pablo Piaggi of Princeton University has managed, with the help of a machine learning model, to simulate the process of ice formation on a computer more accurately than ever before.
How this could improve food processing and also climate models?
Find out in this interview!
- What: Quantum machine learning is the combination of quantum computing power and machine learning technologies. On the one hand, quantum computers can support machine learning, e.g., by speeding up the time it takes to train or evaluate a machine learning model. On the other hand, machine learning methods can be used to develop new quantum algorithms or to uncover codes with which we can correct errors in quantum calculations.
- How: To achieve the former, a hybrid solution is often used: the machine learning algorithm is run on a traditional computer; however, computationally difficult subroutines, which the classical computer would struggle with, are outsourced to a quantum device that can execute more complex calculations faster.
- What for: Quantum machine learning can thus help us solve challenges which require a machine learning approach and also a lot of computing power, like optimizing radiation therapy, predicting the formation of dangerous storms and improving autonomous vehicles.
- Why it’s important now: In only 5 years, quantum computers are supposed to be able to solve relevant real-world problems. Therefore, organizations are well advised to inform themselves about what benefit this can bring their business now.