Our first product is a hardware co-processor called the Optical Processing Unit – or OPU. It is designed to boost some of the most compute-intensive tasks in Machine Learning. The OPU can just be plugged onto a standard server or workstation, and accessed through a simple toolbox that is seamlessly integrated within familiar programming environments. Full-scale OPU prototypes are already available to selected users through the LightOn Cloud. We are opening our registration for those researchers and data scientists interested in trying out our technology on our cloud. The sign-up page is here.
Matrix-vector multiplications are amongst the most important elementary computing blocks in Machine Learning. For instance, Deep Learning schemes essentially stack such matrix-vector multiplications with non-linearities.
An OPU just does that: it multiplies the input data by a fixed matrix, passes through an element-wise non-linearity, and outputs the result. But because the OPU harnesses optics it can do this operation
- at massive data size
- very fast
- at minimal power consumption
What makes each OPU device literally unique is the fixed random matrix at the core if its computations, well fitted to the statistical learning of many Machine Learning / Artificial Intelligence schemes. More technical information on how our Optical Processing Unit perform its operation can be found here.
Examples of successful use cases of the OPU technology include:
- Image and Video Classification
- Recommender Systems
- Anomaly Detection
- Natural Language Processing
Our current Optical Processing Unit performs Random Projections. These Random projections have a long history in terms of permitting the analysis of large sized data. They allow for oblivious dimensionality reduction thanks to the Johnson-Lindenstrauss lemma and have been found useful in large variety of fields such as Compressed Sensing, Randomized Numerical Linear Algebra, Streaming and Sketching algorithms…. Findings using our technology are communicated to the science community through preprints, presentations at conferences, blog posts , workshops and meetups, examples through the use of our API on LightOn Cloud, and publications.
Supervised Learning (Random Features kernels)
- Random Projections through multiple optical scattering: Approximating kernels at the speed of light, Alaa Saade, Francesco Caltagirone, Igor Carron, Laurent Daudet, Angélique Drémeau, Sylvain Gigan, Florent Krzakala
- « Don’t take it lightly: Phasing optical random projections with unknown operators » Sidharth Gupta, Rémi Gribonval, Laurent Daudet, Ivan Dokmanić
- Transfer Learning on the OPU (API example available on LightOn Cloud)
Unsupervised Learning (Randomized Numerical Linear Algebra)
Natural Language Processing
Time series analysis
- High dimensional time series change detection
- « NEWMA: a new method for scalable model-free online change-point detection« , Nicolas Keriven, Damien Garreau, Iacopo Poli
- Special Recurrent Neural Networks
- « Scaling up Echo-State Networks with multiple light scattering« , Jonathan Dong, Sylvain Gigan, Florent Krzakala, Gilles Wainrib, IEEE Statistical Signal Processing Workshop (SSP), Freiburg, Germany, 2018, pp. 448-452
- « Optical Reservoir Computing using multiple light scattering for chaotic systems prediction« , Jonathan Dong, Mushegh Rafayelyan, Florent Krzakala, Sylvain Gigan
Other: Context and General framework linking machine learning and physics (including quantum computing)
- « Machine learning and the physical sciences » Giuseppe Carleo, Ignacio Cirac, Kyle Cranmer, Laurent Daudet, Maria Schuld, Naftali Tishby, Leslie Vogt-Maranto, Lenka Zdeborová.
- LightOn’s Summer Series #1 — Faith No Moore: Silicon Will Not Scale Indefinitely (blog post)
- LightOn’s Summer Series #2 — Optical Computing: a New Hope (blog post)
- Random Projections and the Blessing of Dimensionality (blog post)
- Random Projections at the Speed of Light: Full Ahead Mr. Sulu, Maximum Warp (blog post)