LightOn Optical Processing Unit : Scaling-up AI and HPC with a Non von Neumann co-processor

@article{Brossollet2021LightOnOP,
  title={LightOn Optical Processing Unit : Scaling-up AI and HPC with a Non von Neumann co-processor},
  author={Charles Brossollet and Alessandro Cappelli and Igor Carron and Charidimos Chaintoutis and Am'elie Chatelain and Laurent Daudet and Sylvain Gigan and Daniel Hesslow and Florent Krzakala and Julien Launay and Safa Mokaadi and Fabien Moreau and Kilian Muller and Ruben Ohana and Gustave Pariente and Iacopo Poli and Giuseppe Luca Tommasone},
  journal={2021 IEEE Hot Chips 33 Symposium (HCS)},
  year={2021},
  pages={1-11}
}
Beyond pure Von Neumann processing Scalability of AI / HPC models is limited by the Von Neumann bottleneck for accessing massive amounts of memory, driving up power consumption. 

Figures from this paper

Photonic co-processors in HPC: Using LightOn OPUs for Randomized Numerical Linear Algebra

TLDR
Lighton OPU: The first commercially available photonic Co-Processor is presented, which aims to get approximate results at reduced computational cost by leveraging randomization.

Is the Number of Trainable Parameters All That Actually Matters?

TLDR
This work emulates an increase in effective parameters, usingcient approxi-mations: either by doping the models with frozen random parameters, or by using fast structured transforms in place of dense linear layers, and concludes that the scaling relationship between test loss and compute depends only on the actual number of trainable parameters.

References

SHOWING 1-10 OF 16 REFERENCES

Photonic co-processors in HPC: Using LightOn OPUs for Randomized Numerical Linear Algebra

TLDR
Lighton OPU: The first commercially available photonic Co-Processor is presented, which aims to get approximate results at reduced computational cost by leveraging randomization.

Silicon Photonics for Artificial Intelligence Acceleration : HotChips 32

  • Carl Ramey
  • Art
    2020 IEEE Hot Chips 32 Symposium (HCS)
  • 2020
This article consists only of a collection of slides from the author's conference presentation.

Light-in-the-loop: using a photonics co-processor for scalable training of neural networks

TLDR
This study presents the first optical co-processor able to accelerate the training phase of digitally-implemented neural networks, relying on direct feedback alignment as an alternative to backpropagation, and performs the error projection step optically.

Training of Photonic Neural Networks through In Situ Backpropagation

TLDR
A protocol for training photonic neural networks based on adjoint methods by physically backpropagating an optical error signal and calculating the gradient of the network with respect to its tunable degrees of freedom.

Random projections through multiple optical scattering: Approximating Kernels at the speed of light

TLDR
This work proposes an analog, optical device that performs the random projections literally at the speed of light without having to store any matrix in memory, and shows that, on the MNIST database, the experimental results closely match the theoretical performance of the corresponding kernel.

Kernel Computations from Large-Scale Random Features Obtained by Optical Processing Units

  • Ruben OhanaJonas Wacker L. Daudet
  • Computer Science
    ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
  • 2020
TLDR
Experiments demonstrate that the OPU kernel and its RF approximation achieve competitive performance in applications using kernel ridge regression and transfer learning for image classification, and with time and energy savings.

NEWMA: A New Method for Scalable Model-Free Online Change-Point Detection

TLDR
This paper proposes a new, simple method for model-free online change-point detection that relies only on fast and light recursive statistics, inspired by the classical Exponential Weighted Moving Average algorithm (EWMA).

Adversarial Robustness by Design through Analog Computing and Synthetic Gradients

TLDR
A new defense mechanism against adversarial at-tacks inspired by an optical co-processor is proposed, providing robustness without compromising natural accuracy in both white-box and black-box settings, and the same approach is suboptimal if employed to generate adversarial examples.

Online Change Point Detection in Molecular Dynamics With Optical Random Features

TLDR
This work shows that the No-prior-Knowledge Exponential Weighted Moving Average (NEWMA) algorithm can be used along optical hardware to successfully identify changes in real-time, and may enhance the sampling of the conformational space of molecules.

Fast Graph Kernel with Optical Random Features

TLDR
This paper proposes to leverage kernel random features within the graphlet framework, and establish a theoretical link with a mean kernel metric, and incorporates optical random features that can be computed in constant time.