Alumni politecnico di milano
The cover story in the April edition of Nature Photonics talks about at a study resulting from a partnership between the Politecnico di Milano, CNR (Italian National Research Council) and the University of Vienna. Cover stories don't happen every day; and indeed, it is all about the achievement of something never done before. The papers are calling it the first “artificial quantum neuron” Ansa and RaiNews). are writing about it, for example). It promises to become the missing link between quantum computing and artificial intelligence. Why is it so important? What can it be used for?
We asked Andrea Crespi, researcher at the Politecnico di Milano and Engineering Physics PhD alumna, and member of the team that developed it.
Some of the more efficient “traditional” neural networks, on which modern artificial intelligence algorithms are based, are made up of connections between “artificial neurons” called memristors (memory-resistors). These are electronic components that change their electrical resistance on the basis of a memory of the current that has passed through them previously. Artificial neuron networks can be “taught” to learn thanks to this property, and this is how they learn to perform complex tasks, such as understanding speech, recognising a face, interpreting images (for diagnostic purposes, for example) or driving a car (including racing cars, as achieved by the Polimove team, which programmed the fastest autonomous racing car in the world). It is the technology behind artificial intelligence.
Up to now, we have been talking about electronic devices. However, in the meantime, the scientific world has also developed a new generation of computing, namely quantum computing. The fundamental difference is that instead of using electronic impulses, quantum computing exploits individual particles to code information. For example, Optical impulses made up of individual photons, which behave differently from electric current. The result is a processing power that is potentially much higher than that of the best “traditional” (or electronic) supercomputers.
Since the advantage of quantum computing is that it proportionally increases the number of operations that can be performed, it is particularly efficient for problems that, with an electronic device, would require a vast number of operations (and therefore huge amounts of time and energy) to be solved. Examples of its application include cryptography, search algorithms and physical system simulations.
The concept of quantum computing is nothing new; however, to date, no true quantum neural network has been created. In fact, a fundamental link was missing: the quantum memristor, the artificial quantum neuron. “The idea has existed for several years, but it was only recently that a group of physicians from the University of Vienna demonstrated that it can be done,” says Crespi. His research group, led by professor Roberto Osellame, designed and engineered the first true prototype of a quantum memristor, an optical device with the same functional characteristics as a memristor, capable of operating on quantum states of light:
“The information is no longer coded in an electrical signal, as is the case with traditional memristors. It is the optical transmission that varies; that is, the amount of light that can pass through the device, based on the amount of light that passed through it before.”
Before now, a device of this nature had only been theorised. What the Politecnico has created is therefore the first prototype quantum memristor, and perhaps the first “neuron” in an artificial quantum network.
Credits home and header: Equinox graphics
