Sunrise and Landrush, page-133

One thing I noticed and it was reinforced in the Rob Telson podcast was that @uiux trained the neutral background then placed the Coke can in that space and labelled or trained the image. At that point the only things in view were the neutral background and the Coke can.

Then as the Coke can was returned AKIDA was recognising it even though it was moving and being partially obscured by the hand of the person placing it back into the neutral background space.

During Rob Telson’s interview he refers to being shown one shot learning and refers to asking if it could identify a baseball. They show the baseball to AKIDA and train it with one shot and it identified the baseball then as they threw the baseball across the area being monitored by the camera and it identified it successfully. Assuming it was a normal baseball it would have had stitched seams in a different colour to the leather and like every ball I have ever thrown would have had some rotation as well as it’s forward velocity yet AKIDA still identified it.

Now why did I find this significant well go back to old fashioned machine learning used by Tesla and the poor woman pushing a bike. Old fashioned machine learning did not recognise her because it was confused by her pushing a bike. It needed to be trained with more images which included this scenario.

AKIDA recognised the Coke can even though in the hand of a person and it was moving.

AKIDA recognised the baseball even though it was moving and rotating.

AKIDA2000 with LSTM is adding prediction.

Does anyone now not understand why Peter van der Made is able to say that Brainchip is three years ahead.

My opinion only DYOR
FF

AKIDA BALLISTA