Handwritten Digits Recognition Based on a Parallel Optoelectronic Time-Delay Reservoir Computing System

Abstract

In this work, the performance of an optoelectronic time-delay reservoir computing system for performing a handwritten digit recognition task is numerically investigated, and a scheme to improve the recognition speed using multiple parallel reservoirs is proposed. By comparing four image injection methods based on a single time-delay reservoir, we find that when injecting the histograms of oriented gradient (HOG) features of the digit image, the accuracy rate (AR) is relatively high and is less affected by the offset phase. To improve the recognition speed, we construct a parallel time-delay reservoir system including multi-reservoirs, where each reservoir processes part of the HOG features of one image. Based on 6 parallel reservoirs with each reservoir possessing 100 virtual nodes, the AR can reach about 97.8%, and the reservoir processing speed can reach about 1 × 106 digits per second. Meanwhile, the parallel reservoir system shows strong robustness to the parameter mismatch between multi-reservoirs.

Performance analysis of VCSEL-based reservoir computing system under arbitrary-polarization optical feedback

“…Testing refers to inferring whether the label of the test sample is consistent with the actual label based on the collected test state and the calculated output weight when the test sample is input. Figure 1(b) is the scheme of RC utilizing a single node with time delayed feedback [3]. The major difference between this delay-based reservoir system and Figure 1(a) lies in the second part.…”

“…RC has advantage of easy training, though it originates from recurrent neural networks. Generally, traditional RC needs a huge number of physical nodes to form reservoir to convert the input signal to the reservoir states [1,2]. However, due to the huge number of internal nonlinear nodes, traditional RC has great difficulty in implementation.…”

“…However, due to the huge number of internal nonlinear nodes, traditional RC has great difficulty in implementation. In 2011, Appeltant et al successfully replace the nonlinear nodes connected in the entire network with a single physical node under time-delay feedback [3]. Such a time-delay RC system successfully solves the difficulty of traditional RC in hardware implementation.…”

Photonic Reservoir Computing System for Pattern Recognition Based on an Array of Four Distributed Feedback Lasers

“…In 2021, Yue et al implemented a parallel delayed RC system based on two SLs using electronic injection technology, with an NMSE value below 0.03 and a data processing rate of 28 GSa/s . In 2023, Hou et al proposed the parallel reservoirs of the multiple RC used for handwritten digit recognition tasks, with a recognition accuracy of 0.978 . Subsequently, Zhong et al have demonstrated a P-RC system with three laterally coupled lasers that could reproduce the nonlinear dynamics In 2023, Zou et al proposed a nonfeedback method that utilized the pulse broadening effect induced by optical dispersion to implement a reservior layer by computing the multiplication of the modulator with the summation of the pulse temporal integration of the distributed feedback-laser diode.…”

“…36 In 2023, Hou et al proposed the parallel reservoirs of the multiple RC used for handwritten digit recognition tasks, with a recognition accuracy of 0.978. 37 Subsequently, Zhong et al have demonstrated a P-RC system with three laterally coupled lasers that could reproduce the nonlinear dynamics 38 In 2023, Zou et al proposed a nonfeedback method that utilized the pulse broadening effect induced by optical dispersion to implement a reservior layer by computing the multiplication of the modulator with the summation of the pulse temporal integration of the distributed feedback-laser diode. Their proposed fully analog feed-forward photonic RC (FF-PhRC) system has been experimentally demonstrated to be effective in chaotic signal prediction, spoken digit recognition, and MNIST classification.…”