A team of international researchers has shown how training neural networks directly with light on photonic chips could make AI faster and more sustainable.
A breakthrough study, published in Nature, involved collaboration between the Politecnico di Milano, EPFL Lausanne, Stanford University, the University of Cambridge, and the Max Planck Institute.
The research highlights how physical neural networks, which use analogue circuits that exploit the laws of physics, can process information in new ways.
Photonic chips developed at the Politecnico di Milano perform mathematical operations such as addition and multiplication through light interference on silicon microchips only a few millimetres in size.
By eliminating the need to digitise information, these chips dramatically cut both processing time and energy use. Researchers have also pioneered an ‘in-situ’ training technique that enables photonic neural networks to learn tasks entirely through light signals, instead of relying on digital models.
The result is a training process that is faster, more efficient and more robust.
Such advances could lead to more powerful AI models capable of running directly on devices instead of being dependent on energy-hungry data centres.
An approach that paves the way for technologies such as autonomous vehicles, portable intelligent sensors and real-time data processing systems that are both greener and quicker.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
