Optical Implementation Of Neural Networks Research Articles

Design of Mode-Locked Fibre Laser with Non-Linear Power and Spectrum Width Transfer Functions with a Power Threshold

There is a growing demand for higher computational speed and energy efficiency of machine learning approaches and, in particular, neural networks. Optical implementation of neural networks can address this challenge. Compared to other neuromorphic platforms, fibre-based technologies can unlock a wide bandwidth window and offer flexibility in dimensionality and complexity. Moreover, fibre represents a well-studied, low-cost and low-loss material, widely used for signal processing and transmission. At the same time, mode-locked fibre lasers offer flexibility and control, while the mode-locking effect can be crucial for unlocking ultra-short timescales and providing ultra-fast processing. Here, we propose a mode-locked fibre laser with a non-linear power threshold in both power and spectrum. The advantage of the proposed system is a spectrum width two-branch function dependent on the input signal power. The effect is caused by a transition between two operating regimes and is governed by the input signal power. The proposed design enables receiving a non-linear transfer function in amplitude with a power threshold as an optical analogue of biological neurons with the additional advantage of a non-linear two-branch transfer function in spectrum width. The latter property is similar to the frequency-varied response dependent on stimulus properties in biological neurons. Thus, our work opens new avenues in research into novel types of artificial neurons with a frequency spectrum width variable response and, consequently, spiking neural networks and neural-rate-based coding with potential applications in optical communications and networks with flexible bandwidth, such as 5G and emerging 6G.

Optical coherent dot-product chip for sophisticated deep learning regression

Optical implementations of neural networks (ONNs) herald the next-generation high-speed and energy-efficient deep learning computing by harnessing the technical advantages of large bandwidth and high parallelism of optics. However, due to the problems of the incomplete numerical domain, limited hardware scale, or inadequate numerical accuracy, the majority of existing ONNs were studied for basic classification tasks. Given that regression is a fundamental form of deep learning and accounts for a large part of current artificial intelligence applications, it is necessary to master deep learning regression for further development and deployment of ONNs. Here, we demonstrate a silicon-based optical coherent dot-product chip (OCDC) capable of completing deep learning regression tasks. The OCDC adopts optical fields to carry out operations in the complete real-value domain instead of in only the positive domain. Via reusing, a single chip conducts matrix multiplications and convolutions in neural networks of any complexity. Also, hardware deviations are compensated via in-situ backpropagation control provided the simplicity of chip architecture. Therefore, the OCDC meets the requirements for sophisticated regression tasks and we successfully demonstrate a representative neural network, the AUTOMAP (a cutting-edge neural network model for image reconstruction). The quality of reconstructed images by the OCDC and a 32-bit digital computer is comparable. To the best of our knowledge, there is no precedent of performing such state-of-the-art regression tasks on ONN chips. It is anticipated that the OCDC can promote the novel accomplishment of ONNs in modern AI applications including autonomous driving, natural language processing, and scientific study.

Optical implementation of neural networks for face recognition by the use of nonlinear joint transform correlators

We describe a nonlinear joint transform correlator-based two-layer neural network that uses a supervised learning algorithm for real-time face recognition. The system is trained with a sequence of facial images and is able to classify an input face image in real time. Computer simulations and optical experimental results are presented. The processor can be manufactured into a compact low-cost optoelectronic system. The use of the nonlinear joint transform correlator provides good noise robustness and good image discrimination.

Controlling the grey level capacity of a bistable FLC spatial light modulator

Abstract Due to their memory, non-linear response and threshold, as well as their grey level capacity, ferroelectric liquid crystal spatial light modulators are useful devices for optical implementations of neural networks. We propose a phenomenological explanation of the roles, the different driving parameters like the voltage, the current and the length of the addressing pulse play in achieving and controlling these grey levels. Finally we discuss a possible optical implementation of a self organizing memory based on the model of the Kohonen map, using optically addressed SLMs.

Variable-sensitivity photodetector that uses a metal–semiconductor–metal structure for optical neural networks

A novel type of photodetector called a variable-sensitivity photodetector has been developed for optical implementation of neural networks. It utilizes a metal-semiconductor-metal structure whose quantum efficiency can be modulated by an applied bias voltage. A linear dependence of the sensitivity on the bias voltage was obtained with the bipolar current flow. This device operated as a multiplier of the incident light intensity and the bias voltage. It is shown that this device is suitable for achieving dynamic synaptic interconnections. A 4 x 4 array device was fabricated and demonstrated.

Character recognition using a dynamic optoelectronic neural network with unipolar binary weights

A novel type of quantized learning rule with unipolar binary weights that is useful for the optical implementation of neural networks is reported. An input-dependent thresholding operation is also proposed to remove the unwanted effects that are due to the insufficient contrast ratio of spatial light modulators as synaptic connection devices. Moreover, we experimentally demonstrate the recognition of 26 characters of the alphabet by using the single set of an optoelectronic three-layered network.

Frequency multiplexed raster neural networks 1: Theory

In the first part of this paper we present frequency multiplexed raster (FMR) optical implementation of neural networks. A hidden difficulty for hardware (optical and electronic) implementation is that the dimension of the synaptic matrix is twice that of the input and output matrices or vectors. For 2-D images, which is we believe one of the greatest potentialities of neural networks, the synaptic matrix is 4-D and cannot be directly implemented in optics. We propose FMR as a method to fold this matrix into a 2-D format. In the second part of this paper we describe the system built in our laboratory showing the feasibility of FMR optical neural networks. The system is built from an optical input module, a fixed synaptic matrix coded on a transparency, a CCD camera, and a microcomputer which performs the thresholding and feedback operations. In a later stage the fixed matrix will be replaced by a programmable matrix.