Quantum machine learning: a new tool in the cybersecurity locker

As quantum computing becomes a reality, we are witnessing the formation of the quantum economy. Several companies are already offering quantum-as-a-service and quantum-in-the cloud. However, the ecosystem will soon include many additional services, such as quantum circuit optimization or efficiency advisers. We will see business models and industries, as well as secondary technologies and product offerings, built on this foundation. Prices will go down and availability will go up, including a reduction of entry barriers.

Quantum computing will have a significant impact on several industries – including finance, where institutions will enhance their current financial predictions with quantum technologies; and pharma, where drug discovery and optimization will be enhanced by performing simulations with quantum computers. Quantum computing is a new paradigm, and it represents a revolutionary way of asking questions and building solutions. Many new approaches will be built on it, and it will drive vast changes in many fields.

For several reasons, the combination of quantum computing with machine learning (ML), quantum machine learning (QML), has a very high potential. Both areas have uncertainty at their core, a disadvantage that turns into a strength in the quantum arena. The internal inaccuracy of machine learning enables the calculation of otherwise unattainable outcomes in terms of input volume and calculation speed – in addition to a field of application that exhibits high levels of ambiguity (for example, in image classification, the concepts of “cat” and “dog” are not well-defined). The probabilistic nature of quantum computing, which is a critical impediment to many algorithms, harmonizes well with these properties. For these reasons, QML is on a path to becoming one of the first applications of quantum computing outside academia. As a result, stakeholders in academia, industry and government are already showing considerable interest in quantum machine learning.

Cybersecurity is weak across our digital ecosystem. Companies and governments already face many threats and are struggling to keep their computing environments secure. Conceptually, these threats include well-funded and deeply motivated malicious actors, crime-as-a-service, as well as state-sponsored activity – including, for example, terrorism, cyber warfare and industrial espionage.

Quantum computing has potential to worsen the situation. Adding quantum computing capabilities to the threat could make malicious actors even more effective. For example, new computational options offered by quantum computing could make it easier to find exploitable security gaps or trick existing ML and QML models into a behaviour desired by the attacker.

Yet, the picture is not entirely bleak. Artificial intelligence and machine learning have been used for some time to achieve cybersecurity goals and to protect against threats. ML can detect abnormal behaviour or identify malicious emails or executable code. Most of the time, these applications require huge amounts of data during the training phase. As a result, traditional ML for cybersecurity is very resource-intensive and expensive. However, innovations over the last decade, whether in hardware (like tensor processing units) or in AI models (like ChatAI and ImageAI in various forms and flavours), have lowered the cost of ML, thereby making the development of market products more sustainable.

Quantum machine learning has the potential to generate significant efficiency benefits to the model training phase, making ML a more effective security tool. Due to inherent aspects of quantum computing like entanglement and superposition, the processing of huge datasets can be facilitated and simplified. While the still unsolved issue of data encoding and loading into a quantum computer stands in the way of processing high volumes of data with a quantum computer at this time, we fully expect industry to overcome these challenges over the next decade.

In academia, cybersecurity QML has progressed based on simulations. However, at some point simulations fall short and real-world tests are needed. The detection of spam in emails can be considered a benchmark: it is well understood, not overly complex, and has a vast amount of useful training data available. Together with Fraunhofer IAIS, Capgemini has used an actual quantum computer to conduct spam filtering. While this method is not currently cost-effective, it demonstrates what is possible, as well as the advantages and disadvantages of using today’s quantum devices for cybersecurity QML.

We can foresee other uses for quantum machine learning in cybersecurity, such as mapping critical infrastructure or the cybercriminal ecosystem. Combining these applications will make QML into a highly effective tool for defenders.

Given the current state of quantum computing, cybersecurity companies do not need to create concrete plans for integrating quantum computing into everyday business operations in the near future. That said, the quantum computing ecosystem is evolving, with significant recent advancements (such as Google’s presentation of the first quantum error correction algorithm) and products (such as IBM’s Osprey chip with 433 qubits). At the same time, research projects are breaking the barrier from purely theoretical to simulated and now to actual quantum experiments. Not only do these changes mean that we can reduce the amount of quantum computing hype, but also that we will soon have reliable data points to project what the near future will hold for quantum computing.

Investing in quantum computing requires long-term thinking; tangible returns will not be realized in the near future. Waiting for others to carry the burden, however, brings a risk of being under-prepared and behind the curve. The impending revolution is likely to leave unprepared organizations razed in its wake.

In conclusion, change is coming to the cybersecurity industry that reflects changes in the broader digital ecosystem. While we cannot say precisely how QML will be incorporated into cybersecurity operations, we can foresee its utility as a security tool. Given the overall state of cybersecurity, it cannot get here a moment too soon.