The Origin of our pattern recognition technology

      A vision artificial technology based on human visual system


The visual system of primates is remarkable in that it can analyse and extract the most relevant information in its visual field almost instantly: only a hundred milliseconds are needed, and sufficient, to produce a signal characterising the presence of an animal. This property, which is found in humans and apes (on an even more enhanced level), would certainly have been a great asset in terms of survival for the animals that had developed such a visual system through evolution.

To a certain extent, this property is also the starting point of the technology developed by Spikenet. Simon Thorpe and his teams have spent twenty years investigating rapid visual processing, from both an experimental and theoretical point of view. Through this research, they have developed a robust, if rather unusual, theory to explain the phenomenal capacity of the visual system (something that the classical theories of cerebral information processing were hard put to achieve).

The theory is highly original in that its explicit hypothesis is that neuron firing times can be reproduced (the "spike times"), whereas the traditional school of thought has long preferred, and still does, to focus on firing rates (the "spike rate") to explain the coding of information which implies that neuron response times cannot be characterised with any accuracy.

Following twenty years of successful research, this radical change of viewpoint, moving from a "neuro-centric" to a "spiko-centric" focus, led to the creation of Spikenet Technology, an innovative company working in the field of Artificial Vision which uses this theory as the basis for developing its pattern recognition disruptive technology, directly inspired by the latest advances in Neuroscience.

This conceptual breakthrough has led us to model the visual system not on the basis of artificial neurone networks but of asynchronous spiking neural networks, thereby opening the door to new image analysis applications and pertinent solutions to the emerging issues of our increasingly connected world, by endowing machines with the mechanisms of human vision.

1/10