2020 BRN Discussion, page-27085

My wish for everyone who holds shares in BRN is that they will be able to all fly first class on long haul. There is actually no bad first class be it Emirates, British, Qantas, Singapore, Etihad all great.

On a research note I have come across the following research paper and the lead author and one other are together with Professor Iliadis and Dr. Konstantinos Demertzis guest lecturers at the Hellenic Military Academy. There is a clear connection between these gentlemen through their respective work with the military and NATO but if that connection fails as uiux now famously stated all roads lead to JAST:

Time-Multiplexed Spiking Convolutional NeuralNetwork Based on VCSELs for Unsupervised Image Classification

Abstract

In this work, wepresent numerical results concerning a multilayer “deep” photonic spikingconvolutional neural network, arranged so as to tackle a 2D imageclassification task. The spiking neurons used are typical two-sectionquantum-well vertical-cavity surface-emitting lasers that exhibit isomorphicbehavior to biological neurons, such as integrate-and-fire excitability andtiming encoding. The isomorphism of the proposed scheme to biological networksis extended by replicating the retina ganglion cell for contrast detection inthe photonic domain and by utilizing unsupervised spike dependent plasticity asthe main training technique. Finally, in this work we also investigate thepossibility of exploiting the fast carrier dynamics of lasers so as totime-multiplex spatial information and reduce the number of physical neuronsused in the convolutional layers by orders of magnitude. This last featureunlocks new possibilities, where neuron count and processing speed can beinterchanged so as to meet the constraints of different applications.

Keywords: neuromorphic computing; optical neural networks; image classification; machine learning; laser dynamics; semiconductor lasers; VCSEL

4. Conclusions

At a glance, the proposed neuralscheme is an optical adaptation of a SCNN [36], aiming to inherent the performance of its software counterpart and at the same time provide radical new advantages by replacing software functions and nodes with photonic neurons. The resulting PSCNN comprises VCSEL neurons which are arranged in multiple “deep” neural layers. Each layer provides a different operation, ranging from pixel-contrast encoding to spike-latency, spike time-multiplexing and SCNNs for pattern recognition. In our work, the training of the neuromorphic scheme relies on an unsupervised version of STDP, whereas each node’s response was computed through a physically accurate numerical model. Furthermore, in order to address the high neuron count dictated by SCNNs we realized a time-multiplexing strategy, where different pixels of the image are processed by the same physical laser-neurons. This technique allowed the replication of a software based SCNN with 2020 neurons with only 62 laser-nodes and an inference rate of 1.38 M frame/s for 144 pixel images. Furthermore, we generated an artificial set of images depicting numerical digits so as to train/test the classification capabilities of the proposed network. The results confirm that the integrate-and-fire nature of the VCSEL neurons renders our scheme extremely resilient to typical white noise sources (shot, thermal noise), while variations at the mean intensity of pixels affect image contrast and thus impact spike timing, leading to high classification error.

Author Contributions

M.S. developed the numerical model and the neural network, did themajority of the simulations and wrote the manuscript with help from allco-authors. S.D. was responsible for GPU simulations. G.S. and A.B. provideddiscussion/input on numerical modelling and neural structure simulation. C.M.was the initiator of this project and was supervising the work. All authorshave read and agreed to the published version of the manuscript.

Funding

This research was funded by the EU H2020 NEOTERIC project (871330) andthe Hellenic Foundation for Research and Innovation (HFRI) and the GeneralSecretariat for Research and Technology (GSRT), under grant agreement No 2247(NEBULA project).

References:

1. Gautrais, J.; Thorpe, S. Rate coding versus temporal ordercoding: A theoretical approach. Biosystems 1998, 48, 57–65. [Google Scholar] [CrossRef]

2. Masquelier, T.;Guyonneau, R.; Thorpe, S.J. Competitive STDP based spike patternlearning. Neural Comput. 2009, 21,1259–1276. [Google Scholar] [CrossRef] [PubMed]

https://www.mdpi.com/2076-3417/11/4/1383