At the heart of all artificial intelligence applications are mathematical models called neural networks.
The biological structure of the human brain, which is made up of interconnected nodes, inspired these models.
Neural networks can be mathematically trained by tuning their internal structure until they are capable of human-level tasks, such as recognising our faces, interpreting medical images for diagnosis, and even driving our cars, much like our brain learns by constantly rearranging the connections between neurons.
Having integrated devices capable of swiftly and efficiently completing the computations necessary in neural networks has thus become a major academic and corporate research priority.
The development of the memristor in 2008 was one of the key game changers in the area. The memory-resistor, or memristor, is a device that adjusts its resistance based on a memory of a previous current. Scientists noticed very away that the unique behaviour of memristors was remarkably comparable to that of neural synapses (among many other applications). As a result, the memristor has become a crucial component in neuromorphic designs.
Prof. Philip Walther and Dr. Roberto Osellame of the University of Vienna, the National Research Council (CNR), and the Politecnico di Milano have now demonstrated that it is possible to engineer a device that behaves like a memristor while acting on quantum states and being able to encode and transmit quantum information. To put it another way, a quantum memristor. Because the dynamics of a memristor tend to contradict usual quantum behaviour, realising such a device is difficult.
The physicists overcame the difficulty by using single photons, or single quantum particles of light, and using their unique capacity to propagate simultaneously in a superposition of two or more routes. Single photons propagate along waveguides laser-written on a glass substrate and are steered on a superposition of many routes in their experiment. One of these pathways is used to detect the flux of photons passing through the device, and this quantity regulates the transmission on the other output via a sophisticated electrical feedback mechanism, resulting in the desired memristive behaviour. In addition to demonstrating the quantum memristor, the researchers have provided simulations demonstrating that optical networks with quantum memristors can be used to learn on both classical and quantum tasks, implying that quantum memristors could be used to learn on both classical and quantum tasks.
“Unlocking the full potential of quantum resources within artificial intelligence is one of the greatest challenges of the current research in quantum physics and computer science,” says Michele Spagnolo, who is first author of the publication in the journal “Nature Photonics.” The group of Philip Walther of the University of Vienna has also recently demonstrated that robots can learn faster when using quantum resources and borrowing schemes from quantum computation. This new achievement represents one more step towards a future where quantum artificial intelligence become reality.
