Optical information processing: Advances in nanophotonic reservoir computing

Despite ever increasing computational power, recognition and classification problems remain challenging to solve. Recently advances have been made by the introduction of the new concept of reservoir computing. This is a methodology coming from the field of machine learning and neural networks and has been successfully used in several pattern classification problems, like speech and image recognition. The implementations have so far been in software, limiting their speed and power efficiency. Photonics could be an excellent platform for a hardware implementation of this concept because of its inherent parallelism and unique nonlinear behaviour. We propose using a network of coupled Semiconductor Optical Amplifiers (SOA) and show in simulation that it could be used as a reservoir by comparing it on a benchmark speech recognition task to conventional software implementations. In spite of several differences, they perform as good as or better than conventional implementations. Moreover, a photonic implementation offers the promise of massively parallel information processing with low power and high speed. We will also address the role phase plays on the reservoir performance.

Reservoir computing has attracted much attention for its easy training process as well as its ability to deal with temporal data. A reservoir computing system consists of a reservoir part represented as a sparsely connected recurrent neural network and a readout part represented as a simple regression model. In machine learning tasks, the reservoir part is fixed and only the readout part is trained. Although reservoir computing has been mainly applied to time series prediction and recognition, it can be applied to image recognition as well by considering an image data as a sequence of pixel values. However, to achieve a high performance in image recognition with raw image data, a large-scale reservoir including a large number of neurons is required. This is a bottleneck in terms of computer memory and computational cost. To overcome this bottleneck, we propose a new method which combines reservoir computing with untrained convolutional neural networks. We use an untrained convolutional neural network to transform raw image data into a set of smaller feature maps in a preprocessing step of the reservoir computing. We demonstrate that our method achieves a high classification accuracy in an image recognition task with a much smaller number of trainable parameters compared with a previous study.

Reservoir Computing is a novel computing paradigm that uses a nonlinear recurrent dynamical system to carry out information processing. Recent electronic and optoelectronic Reservoir Computers based on an architecture with a single nonlinear node and a delay loop have shown performance on standardized tasks comparable to state-of-the-art digital implementations. Here we report an all-optical implementation of a Reservoir Computer, made of off-the-shelf components for optical telecommunications. It uses the saturation of a semiconductor optical amplifier as nonlinearity. The present work shows that, within the Reservoir Computing paradigm, all-optical computing with state-of-the-art performance is possible.

Reservoir computing is a potential neuromorphic paradigm for promoting future disruptive applications in the era of the Internet of Things, owing to its well-known low training cost and compatibility with hardware. It has been successfully implemented by injecting an input signal into a spatially extended reservoir of nonlinear nodes or a temporally extended reservoir of a delayed feedback system to perform temporal information processing. Here we propose a novel nondelay-based reservoir computer using only a single micromechanical resonator with hybrid nonlinear dynamics that removes the usually required delayed feedback loop. The hybrid nonlinear dynamics of the resonator comprise a transient nonlinear response, and a Duffing nonlinear response is first used for reservoir computing. Due to the richness of this nonlinearity, the usually required delayed feedback loop can be omitted. To further simplify and improve the efficiency of reservoir computing, a self-masking process is utilized in our novel reservoir computer. Specifically, we numerically and experimentally demonstrate its excellent performance, and our system achieves a high recognition accuracy of 93% on a handwritten digit recognition benchmark and a normalized mean square error of 0.051 in a nonlinear autoregressive moving average task, which reveals its memory capacity. Furthermore, it also achieves 97.17 ± 1% accuracy on an actual human motion gesture classification task constructed from a six-axis IMU sensor. These remarkable results verify the feasibility of our system and open up a new pathway for the hardware implementation of reservoir computing.

Artificial neural networks built with optoelectronic synaptic devices have been proven to process visual information effectively. However, it takes great latency time and energy consumption, especially facing dynamic visual information, due to the separated optical sensor, memory, and process unit. Reservoir computing (RC) based on optoelectronic synaptic devices provides an in-sensor RC for processing temporal information efficiently. It achieves efficient computation by sensing and processing optical signals directly with optoelectronic synaptic devices. Optoelectronic synaptic devices shine in visual information processing, whose application in visual sensing and processing will provide a viable hardware solution for in-sensor computing. Therefore, the application of optoelectronic synaptic devices in reservoir computing has prompted increasing attention. Herein, for promoting the application of physical reservoir computing (PRC) with optoelectrical synapses in machine vision, synaptic plasticity will be introduced first in this work and then illustrate the basic functions of optoelectronic synapses as well as their application in reservoir computing further, and provide a perspective on PRC with optoelectronic synapses in the final.

Micro-ring resonator based all-optical Arithmetic and Logical Unit

The aim of this presentation is to show how various ideas coming from the nonlinear stability theory of functional differential systems, stochastic modeling, and machine learning, can be put together in order to create an approximating model that explains the working mechanisms behind a certain type of reservoir computers. Reservoir computing is a recently introduced brain-inspired machine learning paradigm capable of excellent performances in the processing of empirical data. We focus on time-delay based reservoir computers that have been physically implemented using optical and electronic systems and have shown unprecedented data processing rates. Reservoir computing is well-known for the ease of the associated training scheme but also for the problematic sensitivity of its performance to architecture parameters. The reservoir design problem is addressed, which remains the biggest challenge in the applicability of this information processing scheme. Our results use the information available regarding the optimal reservoir working regimes in order to construct a functional link between the reservoir parameters and its performance. This function is used to explore various properties of the device and to choose the optimal reservoir architecture, thus replacing the tedious and time consuming parameter scannings used so far in the literature.

Recent advances in electro-optics have brought into use communication and information theory to analyze performance in coherent and incoherent optical information processing systems. An optical information processsing system can be analyzed with many of the same concepts of linear system theory (e.g., spatial impulse response, spatial frequency and spatial domain synthesis, etc.), and the photographic images to be processed can be regarded in the same manner as time signals (e.g., spatial frequency content, spatial amplitude and phase modulation, space-bandwidth product, etc.). Both coherent and incoherent optical processing systems can be treated as linear systems, and the processing operation can generally be carried out by communication theory concepts. Although coherent optical information processing operations have been used for performing complex amplitude operations, complex processing can also be performed with incoherent or white-light illumination. The importance of optical information processing operations, either coherent or incoherent, is due to the basic Fourier transform properties of lenses. In this paper, we will discuss mostly the incoherent systems because they are of more recent interest and possess certain advantages, we feel, over the traditional coherent optical processors. Experimental illustrations of the results are provided. In view of the broad area in optical processing of photographic images, we will confine ourselves to a few applications that we consider of general interest. We apologize for the omission of other techniques and applications, and for neglecting the inclusion of their references.

We present our latest progress using new neuromorphic paradigms for optical information processing in silicon photonics. We show how passive reservoir computing chips can be used to perform a variety of tasks (bit level tasks, nonlinear dispersion compensation, etc.) at high speeds and low power consumption. In addition, we present a spatial analog of reservoir computing based on pillar scatterers and a cavity that can be used to speed up classification of biological cells.

Physical reservoir computing utilizes a physical system as a computational resource. This nontraditional computing technique can be computationally powerful, without the need of costly training. Here, a Hopf oscillator is implemented as a reservoir computer by using a node-based architecture; however, this implementation does not use delayed feedback lines. This reservoir computer is still powerful, but it is considerably simpler and cheaper to implement as a physical Hopf oscillator. A non-periodic stochastic masking procedure is applied for this reservoir computer following the time multiplexing method. Due to the presence of noise, the Euler–Maruyama method is used to simulate the resulting stochastic differential equations that represent this reservoir computer. An analog electrical circuit is built to implement this Hopf oscillator reservoir computer experimentally. The information processing capability was tested numerically and experimentally by performing logical tasks, emulation tasks, and time series prediction tasks. This reservoir computer has several attractive features, including a simple design that is easy to implement, noise robustness, and a high computational ability for many different benchmark tasks. Since limit cycle oscillators model many physical systems, this architecture could be relatively easily applied in many contexts.

By using nonlinearity and optical feedback the flexibility of optical processing can be increased. Logarithm, equi-density contour mapping, and A/D conversion are nonlinear operations which have already been successfully implemented with halftone screen processes [1–3]. These and other nonlinear operations as well as optical digital processing can be performed by suitable image thresholding and planar logic devices [4,5]. To accomplish optical feedback, partially reflecting mirrors have been utilized [5–7]. Optical processing systems with coherent feedback allow for the realization of the well-known feedback transfer function in two dimensions. In this paper some progress during the last year in the studies at UCSD involving nonlinear optical devices and optical feedback is reported, together with brief reviews of relevant background information.

Reservoir computing is a recently introduced brain-inspired machine learning paradigm capable of excellent performances in the processing of empirical data. We focus in a particular kind of time-delay based reservoir computers that have been physically implemented using optical and electronic systems and have shown unprecedented data processing rates. Reservoir computing is well-known for the ease of the associated training scheme but also for the problematic sensitivity of its performance to architecture parameters. This article addresses the reservoir design problem, which remains the biggest challenge in the applicability of this information processing scheme. More specifically, we use the information available regarding the optimal reservoir working regimes to construct a functional link between the reservoir parameters and its performance. This function is used to explore various properties of the device and to choose the optimal reservoir architecture, thus replacing the tedious and time consuming parameter scannings used so far in the literature.

From logical reasoning to mental simulation, biological and artificial neural systems possess an incredible capacity for computation. Such neural computers offer a fundamentally novel computing paradigm by representing data continuously and processing information in a natively parallel and distributed manner. To harness this computation, prior work has developed extensive training techniques to understand existing neural networks. However, the lack of a concrete and low-level machine code for neural networks precludes us from taking full advantage of a neural computing framework. Here we provide such a machine code along with a programming framework by using a recurrent neural network—a reservoir computer—to decompile, code and compile analogue computations. By decompiling the reservoir’s internal representation and dynamics into an analytic basis of its inputs, we define a low-level neural machine code that we use to program the reservoir to solve complex equations and store chaotic dynamical systems as random-access memory. We further provide a fully distributed neural implementation of software virtualization and logical circuits, and even program a playable game of pong inside of a reservoir computer. Importantly, all of these functions are programmed without requiring any example data or sampling of state space. Finally, we demonstrate that we can accurately decompile the analytic, internal representations of a full-rank reservoir computer that has been conventionally trained using data. Taken together, we define an implementation of neural computation that can both decompile computations from existing neural connectivity and compile distributed programs as new connections.

Photonic reservoir computing has been intensively investigated to solve machine learning tasks effectively. A simple learning procedure of output weights is used for reservoir computing. However, the lack of training of input-node and inter-node connection weights limits the performance of reservoir computing. The use of multiple reservoirs can be a solution to overcome this limitation of reservoir computing. In this study, we investigate parallel and deep configurations of delay-based all-optical reservoir computing using semiconductor lasers with optical feedback by combining multiple reservoirs to improve the performance of reservoir computing. Furthermore, we propose a hybrid configuration to maximize the benefits of parallel and deep reservoirs. We perform the chaotic time-series prediction task, nonlinear channel equalization task, and memory capacity measurement. Then, we compare the performance of single, parallel, deep, and hybrid reservoir configurations. We find that deep reservoirs are suitable for a chaotic time-series prediction task, whereas parallel reservoirs are suitable for a nonlinear channel equalization task. Hybrid reservoirs outperform other configurations for all three tasks. We further optimize the number of reservoirs for each reservoir configuration. Multiple reservoirs show great potential for the improvement of reservoir computing, which in turn can be applied for high-performance edge computing.

The increasing speed of fibre-optic-based telecommunications has focused attention on high-speed optical processing of digital information. Complex optical processing requires a high-density, high-speed, low-power optical memory that can be integrated with planar semiconductor technology for buffering of decisions and telecommunication data. Recently, ring lasers with extremely small size and low operating power have been made, and we demonstrate here a memory element constructed by interconnecting these microscopic lasers. Our device occupies an area of 18 x 40 microm2 on an InP/InGaAsP photonic integrated circuit, and switches within 20 ps with 5.5 fJ optical switching energy. Simulations show that the element has the potential for much smaller dimensions and switching times. Large numbers of such memory elements can be densely integrated and interconnected on a photonic integrated circuit: fast digital optical information processing systems employing large-scale integration should now be viable.