To speed up initial learning, the process is to learn the...

To speed up initial learning, the process is to learn the background first so that when you introduce an object you want to learn, AKIDA recognises the whole object as the object it needs to learn affectively subtracting the learned background from the new image.
In the demo the background was learned and then the background was accidentally changed and a new object (tiger) was introduced, which is going to introduce errors.
When don correctly, once the object is learned the background can change all the time as AKIDA will recognise the object it learned against any background.
Becouse Akida does incremental learning even the mistake that was made will eventually correct itself as it sees the tiger in different backgrounds in deferent angles
So your preserved problem with changing background, is actually not a problem.