Artificial intelligence algorithms are based on mathematical models called neural networks, inspired by the biological structure of the human brain, which is made up of interconnected nodes (neurons). Just as in our brain the learning process is based on the rearrangement of the connections between neurons, artificial neural networks can be "trained" on a set of known data that modify its internal structure, making it capable of performing "human" tasks, such as face recognition, the interpretation of medical images to diagnose diseases and even driving a car.
For this reason, research is underway, at academic and industrial level, aiming to obtain integrated and compact devices capable of performing the mathematical operations required for the operation of neural networks in a rapid and efficient way.
A breakthrough in this field was the discovery of the memory-resistor or memristor, a component that changes its electrical resistance based on a memory of the current that passed through it. Scientists have realized that this functioning is surprisingly similar to that of neural synapses, i.e. the connections between neurons in the brain, and the memristor has become a fundamental component with which to build neuromorphic architectures, that is, forged as a model of our brain.
A group of experimental physicists led by Roberto Osellame, research director at the Institute of Photonics and Nanotechnologies of the National Research Council (CNR-IFN), and Philip Walther, professor at the University of Vienna, in collaboration with Andrea Crespi, Associate Professor at the Politecnico di Milano, have shown that it is possible to engineer an optical device with the same functional characteristics as the memristor, capable of operating on quantum states of light and thus encoding and transmitting quantum information: a quantum memristor.
"Making such a device is no trivial matter, since the dynamics of the memristor tend to compromise certain advantageous aspects of quantum devices. Our researchers have overcome this challenge by employing single photons (single particles of light) and exploiting their quantum ability to propagate simultaneously in two or more paths," explains Osellame.
"These photons are conducted in what are known as optical circuits, fabricated by means of laser pulses in a glass chip, dynamically reconfigurable, which can support quantum states of superposition on different paths. By measuring the flow of photons propagating on one of these paths, it is possible, through a complex scheme of electronic feedback, to reconfigure the transmission of the device on the other output, and this enables us to obtain a functionality equivalent to that of the memristor."
"We also simulated an entire optical network made up of quantum memristors," explains Andrea Crespi, "showing that it could be used to learn both classical and quantum tasks". This result seems to suggest that the quantum memristor may be the missing link between artificial intelligence and quantum computing.
"Unleashing the potential of quantum resources within artificial intelligence applications is one of the greatest challenges of current research, both in quantum physics and in computer science," concludes Michele Spagnolo, from the University of Vienna and first author of the scientific publication scientific publication that received the cover of the April issue of Nature Photonics magazine and a comment in the "News & Views" of the same issue. These new results are a step forward towards a future in which quantum artificial intelligence will be a reality.