Loading...

Novel Optical Implementations of Reservoir Computing with Single or Limited Number of Neurons

Boshgazi, Somayeh | 2019

1576 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 52449 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Mehrany, Khashayar; Memarian, Mohammad
  7. Abstract:
  8. Artificial neural networks are systems based on the brain’s functionality which in many cases are able to process highly complex computational tasks like speech recognition, image recognition ,and time series prediction. Due to the complexity of training algorithms in recurrent neural networks, reservoir computers have significant importance in machine learning.Due to low power consumption and crosstalk, high bandwidth and high-speed computing in optics, reservoir computing has proceeded to optical implementations. A reservoir computer consists of three layers: the input layer, reservoir, and output layer. A recurrent neural network is usually used as the reservoir in reservoir computers. Implementation of neurons in the reservoir layer can be physical or virtual. Physical implementations of reservoir computers usually use a large number of neurons. Thus, a recent idea in implementing reservoir computers is the use of time multiplexing approach and one nonlinear node and a delay line, in which neurons are connected virtually in the time domain. In hardware implementations of reservoir using this time multiplexing approach, the delay line is implemented by a long length of optical fiber, which is an obstacle to integration. Therefore, the main purpose of this thesis is to find a suitable alternative to an all-optical integrated circuit for reservoir computing. Our idea is to use a resonator instead of an optical fiber. The use of resonator causes fundamental differences in the architecture of the virtual neural network. In this thesis, we will investigate the differences between these two approaches and show the performance of our proposed structure in some standard tasks. The proposed architecture is shown to handle standard tasks while being much smaller
  9. Keywords:
  10. Optical Reservoir Computing ; Optical Neural Networks ; Reservoir Optical Implementation ; Integrated Neural Network ; Optical Resonators

 Digital Object List

 Bookmark

...see more