2021 BRN Discussion, page-2161

Optimized spiking neurons can classify images withhigh accuracy through temporal coding with two spikes

• Christoph Stöckl &
• Wolfgang Maass

NatureMachine Intelligence volume 3, pages230–238(2021)Cite this article

• 200 Accesses

• Metricsdetails

Abstract

Spike-basedneuromorphic hardware promises to reduce the energy consumption of imageclassification and other deep-learning applications, particularly on mobilephones and other edge devices. However, direct training of deep spiking neuralnetworks is difficult, and previous methods for converting trained artificialneural networks to spiking neurons were inefficient because the neurons had toemit too many spikes. We show that a substantially more efficient conversionarises when one optimizes the spiking neuron model for that purpose, so that itnot only matters for information transmission how many spikes a neuron emits,but also when it emits those spikes. This advances the accuracy that can beachieved for image classification with spiking neurons, and the resultingnetworks need on average just two spikes per neuron for classifying an image.In addition, our new conversion method improves latency and throughput of theresulting spiking networks.

https://www.nature.com/articles/s42256-021-00311-4

This article which was published on 11 March, 2021 refers to the type of method that Peter van der Made and colleagues have applied to patent in the recent post by @BarrelSitter. The amazing thing about what Peter van der Made and colleagues have achieved is that they have reduced the number of spikes in their version to one spike for classifying an image by utilising SRAM to store the spikes until recognition is achieved assuming I have understood this part of the process correctly. By doing this their system will only continue to store spikes until it is satisfied that enough spikes have been received to decide what the image is of and then it makes the classification.

They continue to be way ahead of the curve in my opinion.

My opinion only DYOR.