● QML has the potential to revolutionize cybersecurity and the detection of threats (such as malware) and to facilitate faster and more cost-effective AI training that makes use of smaller datasets.
● In the future, cybersecurity may soon be enhanced by hybrid systems that combine conventional ML and QML, explains Grégoire Barrué, a researcher in AI and Quantum AI for Cybersecurity at Orange.
What is the fundamental difference between traditional machine learning (ML) and quantum machine learning?
Quantum machine learning and traditional ML share the same core principles: the use of a model with parameters that are optimized through training to generate predictions on new data. The fundamental difference lies in the underlying hardware architecture and the computational logic. In quantum machine learning, data is encoded into quantum states, which serve as the input. Furthermore, quantum models, which are specifically designed for quantum hardware, require a completely different process for development and coding.
QML can re-inject (or ‘infuse’) input data throughout the learning process without degrading feature representations learned in preceding layers
Does quantum machine learning perform better? Is QML more effective than classic ML?
Research in quantum machine learning is ongoing, and we are still working on the identification of real-world use cases where it will be particularly effective. It has several potential advantages, notably the ability to train models with significantly smaller datasets: QML can re-inject (or ‘infuse’) input data throughout the learning process — between each quantum neural layer — without degrading feature representations learned in preceding layers. In other words, there is less information loss within quantum models. It follows that quantum machine learning is particularly effective in data-scarce environments and where data is evolving very rapidly.
What about power efficiency and other limitations?
In contrast to classical generative AI, which involves massive energy overheads, most quantum computers — particularly those based on photonic technology — use much less power. Some quantum computers, such as those made by Quandela, only consume 5kW and can be plugged into standard power outlets. However, in view of the complexity of the underlying mathematics and physics, the development of quantum algorithms remains challenging. For instance, gradient descent — a foundational method for training classical neural networks — is difficult to implement in quantum frameworks. QML necessitates a complete redesign of optimization algorithms, which requires specialized expertise in quantum mechanics and computation.
Your work specifically focuses on cybersecurity. What role can quantum machine learning play in this field, notably with regard to threats such as malware?
In a context where classical models are becoming less robust, QML offers a number of advantages. The advent of generative AI has made it much easier to produce malware, which can proliferate very rapidly, and it is also growing more difficult to detect.
The effectiveness of classical models depends on increasingly frequent, and expensive retraining, while QML can achieve comparable or superior detection performance with smaller datasets and lightweight but highly expressive quantum algorithms, which significantly reduce the temporal and financial costs of model optimization.
Will AI and cybersecurity be 100% quantum in the future?
It is still premature to claim that the future will be entirely quantum. Current trends point instead toward the development of hybrid algorithms that leverage the combined strengths of both classical and quantum computing. The advent of quantum processing units (QPUs) is analogous to the historical emergence of GPUs: there are applications where they are useful, and others where classical computing is already sufficient.
How is France positioned with regard to quantum computing?
France is neither ahead nor behind in quantum computing. The desire to avoid being a latecomer as was the case with AI has created a momentum. Events like the QUEST-IS conference have highlighted a new trend for quantum computing in France and Europe. Numerous companies (hardware builders and cloud service providers like OVH) are embarking on initiatives to develop in-house expertise, while universities are offering training programs. At the same time, however, expertise remains concentrated within small groups inside these organizations.
Grégoire Barrué
