Condensed and Living Matter Seminar: Optical neural networks for faster AI and super resolution imaging

Alex Lvovsky (University of Oxford)
- | David Rittenhouse Laboratory, A8
Prof Alexander Lvovsky 0003r_1.jpg

Although machine intelligence is taking over the world, its current digital electronic platform is very inefficient in terms of energy consumption. Switching to analogue computation, which function more like human brains than digital computers, will allow enhancing the energy efficiency by several orders of magnitude. Optics presents a particularly promising platform for analogue AI; however, significant challenges – particularly in the domain of neural network training – must be overcome before it can compete with its digital counterpart. A likely upcoming range of applications of optical neuron networks is in computer vision, as they will allow eliminating the bottleneck associated with back-and-forth conversion of data between optical and electronic formats. A further benefit of optical processing is enhancing the quality of imaging. For example, it allows reaching the quantum frontier of imaging resolution beyond Rayleigh’s diffractive limit which applies to most of the modern classical imaging technology.

Biography: see https://www.physics.ox.ac.uk/our-people/lvovsky.