Is it possible to do predictions with TM backwards ? AND of course mid-sequence too !
If not, will it work if I train with the sequence and then also with its reverse ?
Hey =)
I once tried it and for me it didn’t work… I think it is because of the dependence of HTM on real streaming data… but this is just an idea 
In my case it was a cross validation… and was working for all forward streams… but as soon as I passed the backward ones there were a lot of errors 
But maybe someone else knows more details than me =)
1 Like
Backwards will not work unless it has seen backwards sequences before. Playing something backwards is a high-level task.
Mid-sequence will work. You must train with the reverse patterns and treat them just like any other patterns if you want to match them.