Advanced search
1 file | 1.90 MB Add to list

Programmable tanh-, ELU-, sigmoid-, and Sin-based nonlinear activation functions for neuromorphic photonics

Author
Organization
Abstract
We demonstrate a programmable analog opto-electronic (OE) circuit that can be configured to provide a range of nonlinear activation functions for incoherent neuromorphic photonic circuits at up to 10 Gbaud line-rates. We present a set of well-known activation functions that are typically used to train DL models including tanh-, sigmoid-, ReLU- and inverted ReLU-like activations, introducing also a series of novel photonic nonlinear functions that are referred to as Rectified Sine Squared (ReSin), Sine Squared with Exponential tail (ExpSin) and Double Sine Squared. Experimental validation for all these activation functions is performed at 10 Gbaud operation. The ability of the mathematically modelled photonic activation functions to train Deep Neural Networks (DNNs) has been verified through their employment in Deep Learning (DL) models for MNIST and CIFAR10 classification purposes, comparing their performance against corresponding NNs that utilize an ideal ReLU activation function.
Keywords
NETWORKS, Nonlinear activation function, opto-electronic activation function, opto-electro-optic activation function, programmable analog device

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 1.90 MB

Citation

Please use this url to cite or link to this publication:

MLA
Pappas, Christos, et al. “Programmable Tanh-, ELU-, Sigmoid-, and Sin-Based Nonlinear Activation Functions for Neuromorphic Photonics.” IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, vol. 29, no. 6, 2023, doi:10.1109/JSTQE.2023.3277118.
APA
Pappas, C., Kovaios, S., Moralis-Pegios, M., Tsakyridis, A., Giamougiannis, G., Kirtas, M., … Pleros, N. (2023). Programmable tanh-, ELU-, sigmoid-, and Sin-based nonlinear activation functions for neuromorphic photonics. IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, 29(6). https://doi.org/10.1109/JSTQE.2023.3277118
Chicago author-date
Pappas, Christos, Stefanos Kovaios, Miltiadis Moralis-Pegios, Apostolos Tsakyridis, George Giamougiannis, Manos Kirtas, Joris Van Kerrebrouck, et al. 2023. “Programmable Tanh-, ELU-, Sigmoid-, and Sin-Based Nonlinear Activation Functions for Neuromorphic Photonics.” IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS 29 (6). https://doi.org/10.1109/JSTQE.2023.3277118.
Chicago author-date (all authors)
Pappas, Christos, Stefanos Kovaios, Miltiadis Moralis-Pegios, Apostolos Tsakyridis, George Giamougiannis, Manos Kirtas, Joris Van Kerrebrouck, Gertjan Coudyzer, Xin Yin, Nikolaos Passalis, Anastasios Tefas, and Nikos Pleros. 2023. “Programmable Tanh-, ELU-, Sigmoid-, and Sin-Based Nonlinear Activation Functions for Neuromorphic Photonics.” IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS 29 (6). doi:10.1109/JSTQE.2023.3277118.
Vancouver
1.
Pappas C, Kovaios S, Moralis-Pegios M, Tsakyridis A, Giamougiannis G, Kirtas M, et al. Programmable tanh-, ELU-, sigmoid-, and Sin-based nonlinear activation functions for neuromorphic photonics. IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS. 2023;29(6).
IEEE
[1]
C. Pappas et al., “Programmable tanh-, ELU-, sigmoid-, and Sin-based nonlinear activation functions for neuromorphic photonics,” IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, vol. 29, no. 6, 2023.
@article{01H665715H91QCBBKJVKGS89W5,
  abstract     = {{We demonstrate a programmable analog opto-electronic (OE) circuit that can be configured to provide a range of nonlinear activation functions for incoherent neuromorphic photonic circuits at up to 10 Gbaud line-rates. We present a set of well-known activation functions that are typically used to train DL models including tanh-, sigmoid-, ReLU- and inverted ReLU-like activations, introducing also a series of novel photonic nonlinear functions that are referred to as Rectified Sine Squared (ReSin), Sine Squared with Exponential tail (ExpSin) and Double Sine Squared. Experimental validation for all these activation functions is performed at 10 Gbaud operation. The ability of the mathematically modelled photonic activation functions to train Deep Neural Networks (DNNs) has been verified through their employment in Deep Learning (DL) models for MNIST and CIFAR10 classification purposes, comparing their performance against corresponding NNs that utilize an ideal ReLU activation function.}},
  articleno    = {{6101210}},
  author       = {{Pappas, Christos and  Kovaios, Stefanos and  Moralis-Pegios, Miltiadis and  Tsakyridis, Apostolos and  Giamougiannis, George and  Kirtas, Manos and Van Kerrebrouck, Joris and Coudyzer, Gertjan and Yin, Xin and  Passalis, Nikolaos and  Tefas, Anastasios and  Pleros, Nikos}},
  issn         = {{1077-260X}},
  journal      = {{IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS}},
  keywords     = {{NETWORKS,Nonlinear activation function,opto-electronic activation function,opto-electro-optic activation function,programmable analog device}},
  language     = {{eng}},
  number       = {{6}},
  pages        = {{10}},
  title        = {{Programmable tanh-, ELU-, sigmoid-, and Sin-based nonlinear activation functions for neuromorphic photonics}},
  url          = {{http://doi.org/10.1109/JSTQE.2023.3277118}},
  volume       = {{29}},
  year         = {{2023}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: