Optical AI Processor Company Raises $3.3m

VentureBeat: LightOn, a Paris, France-based AI startup, has closed a $3.3M (€2.9M) seed funding round. LightOn is developing a new optics-based data processing technology for AI. Leveraging compressive sensing, LightOn’s hardware and software can make Artificial Intelligence computations both simpler and orders of magnitude more efficient. The technology, licensed by PSL Research University, was originally developed at several of Paris’ leading research institutions.

For the past few months, LightOn has allowed access to their Optical Processing Units (OPU) to a select group of beta customers through the LightOn Cloud, thanks to a partnership with OVH, Europe’s leading cloud provider. First users from both Academia and Industry have already successfully demonstrated impressive results on this hybrid CPU/GPU/OPU server, outperforming silicon-only computing technology in a variety of large scale Machine Learning tasks. Typical use cases currently include transfer learning, change point detection, or time series prediction.

LightOn’s CEO, Igor Carron said, “It’s an exciting time as Artificial Intelligence develops rapidly. The requirements as usage scales necessitate improved power efficiency and performance. LightOn’s technology addresses these monumental challenges.“

LightOn's OPU technology has been presented in 2015 arxiv.org paper "Random Projections through multiple optical scattering: Approximating kernels at the speed of light" by Alaa Saade, Francesco Caltagirone, Igor Carron, Laurent Daudet, Angélique Drémeau, Sylvain Gigan, Florent Krzakala

"Random projections have proven extremely useful in many signal processing and machine learning applications. However, they often require either to store a very large random matrix, or to use a different, structured matrix to reduce the computational and memory costs. Here, we overcome this difficulty by proposing an analog, optical device, that performs the random projections literally at the speed of light without having to store any matrix in memory. This is achieved using the physical properties of multiple coherent scattering of coherent light in random media. We use this device on a simple task of classification with a kernel machine, and we show that, on the MNIST database, the experimental results closely match the theoretical performance of the corresponding kernel. This framework can help make kernel methods practical for applications that have large training sets and/or require real-time prediction. We discuss possible extensions of the method in terms of a class of kernels, speed, memory consumption and different problems."