Paris, February 16th, 2018
By Igor Carron, CEO, LightOn
LightOn develops a light-based technology for Artificial Intelligence at scale. Today, we are proud to announce the successful integration of our latest Optical Processing Unit (OPU) prototype within a data center, through our partnership with OVH.
LightOn's first computational tests performed on that cloud platform confirm the unequalled ability for LightOn's OPU to speed up large scale Machine Learning tasks. For one such task called Transfer Learning, computational time has been reduced from 20 minutes with a state-of-the-art algorithm running on a Graphics Processing Unit (GPU), to about 3.5 minutes running on the OPU. Because it uses light, LightOn's OPU consumes less power (less than 30 Watts) resulting in a 30x energy savings compared to the GPU on this task. Another benchmark on time-series analysis with a Recurrent Neural Network demonstrates a 200x speedup over conventional CPUs with large RAMs.
LightOn's computing technology will be made available to the Machine Learning community through the LightOn Cloud service, opening up for beta-users in the Spring 2018. That service will be accessed through a user interface akin to other well-known cloud platforms, and will provide full access to virtual machines that feature a combination of CPU, GPU and LightOn's OPU. Our API seamlessly integrates the OPU within a standard Python environment. It works with PyTorch, and is already compliant with Scikit-Learn. It will eventually become compatible with other popular Machine Learning frameworks such as TensorFlow / Keras.
Several companies and research centers have already expressed their interest in trying LightOn's OPU technology for their R&D needs. We believe our technology will permit fast iteration in data scientists' workflows and model exploration. Dr Maurizio Filippone, a world expert in computational statistics at Eurecom (France), declares:
"I'm hugely excited by the Optical Processing Units that LightOn is developing, which promise to tackle randomized computations faster than popular Graphical Processing Units, while allowing for larger computations with lower power consumption. This can be a game-changer in our field."
Data scientists and Machine Learning engineers interested in trying out the LightOn Cloud can sign up at: https://goo.gl/beQXZz