I just got temporal working, in my htm, its now pumping out (in feed-BACK mode) 256 frame sections of a little man running and jumping up and down.
the matrices on the screen are the regions, the yellow numbers are how many space types its stored at that node, and the pink number is how many temporal types at that node. A problem I have now is if I let it record for too long, the amount of space types shoots through the roof (past the capacity of the memory) - as the temporal types of the node before are the space types of the node after, so theres a problem there... I need some way to remedy this problem... some kind of temporal pooling?
But its working.
the smaller regions at the top of the heirarchy work in slower time scale than the higherfrequency large layer (at the top of the screen) each region has 4 frames of data, then the region after has 4 frames of those 4 frames, then 4 frames of those, etcetera, ive got it working on a clock, cause it seemed easier this way.
the single node at the end (with 15 types) is basicly 15 different 64 frame movies that have been recorded into it.
Im gonna try inference, and interactivity by getting input information (as a concept persay) injected during feedforward.
I want to try and control the video somehow.
It doesnt have to stop at 1 bit graphics either, I can record rgb hi res graphics and just edge detect and downsample it into the system, ill be seeing how that goes too.