Ann: Unsupervised Visual Learning Achieved, page-79

The focus is on the keyword visual.

An average computer will process an image that a programmer specifies. Then it will go through a range of videos or images to search for the closest match. This process could take a long time from training to pattern recognition.

So they conducted the test on a highway and fed the input sensor (DAVIS) with visuals of the moving cars. In seconds the SNAP processor learned what was happening and started counting cars. It could do this by processing events that were output by the DAVIS camera. This is visual unsupervised learning.

Imagine if a self driving car could visually via some on-board visual sensor detect the distance from a moving car it could avoid collision. The same could be used for drones.

I think this announcement forms the first physical confirmation of unsupervised learning in real world that has been tested by BRN. While the race car was/might have been fed arbitrarily generated inputs this one was done on a real camera and on a road which shows the renewed interest on this stock.

IniLabs technologies are now in use at over 100 labs and companies around the world in industries including aerospace, automotive, consumer electronics, industrial vision and security. This might mean that an existing partner/customer of Inilabs is interested and we are developing/validating this for them.