2020 BRN Discussion, page-18273

Hi Branded

Sensible question but consider this proposition. When a baby is born it does not know anything and during its first 2 years of life acquires a massive amount of data (knowledge) starting with what its mothers face looks like. Once it knows its mothers face it can then recognise its mothers face from thousands of other mothers faces. The thing about babies is they have relatively short vision and when first born their peripheral vision is still developing. So this assists them with learning their mothers face without being confused by changing backgrounds.

So let us go to AKD1000. It does not know what a tiger looks like and so it has to be shown a tiger. The sensor which shows it the tiger however will include the background and if the background contains thousands of different animals and you say to a AKIDA this is a tiger it will not work. You need to provide it with the type of background which the baby's vision gives it to learn who is its mother. Substitute here a lighting box with white walls and a tiger standing in the middle being seen for the first time by AKD1000 the way a baby first sees its mother due to its limited vision at birth.

Once AKD1000 has been taught what a tiger looks like then you can confuse the background with movement (or as the techies like to call it noise) and AKD1000 will be able to isolate and identify the tiger. That is why AKD1000 has an onboard library and memory and why when you read about
AKIDA technology Brainchip will tell you that if you have your own CNN data set AKD1000 can on chip convert that data set from CNN2SNN and store it in memory. AKD1000 is not more than human. It cannot identify and label something which it has not been educated/trained to identify.

So for autonomous driving it will still use data held in its library on chip as well as data that has been stored in its memory to identify those things which the sensor be it lidar, camera, DVS, radar or 3D point passes to it as data or spikes as it has been trained to identify them or their patterns. The fact that the background changes makes no difference when the object is an object that it has been trained to identify. A baby still recognises its mothers face even if the background is changed from white to floral.

My opinion only DYOR.