I’ve just started reading about SMI, and as i understood it roughly uses the same algorithm with TM but the distal context instead of showing predictive cells, encodes the location, which is then intersected with the proximal input that encodes the feature, but i didn’t get how that location is encoded ?
can somebody give me some explanation or some links on it ?
So not exactly. The distal context will still cause cells to become predictive, just like in the TM. It’s just that the HTM neurons in the layer are no longer connecting to each other’s distal dendrites. They are all getting their distal input from outside the layer. The layer doesn’t know where it’s getting the input, or what it represents. This is one of the problems the layer solves. It must learn the patterns from scratch.
So we can call this “the location signal” and ignore the fact that we don’t know where it comes from or how it is encoded. There is a section about this in the paper.
That being said, we’re trying to figure out how and where this location signal is generated.
by that you mean that the location signal actually “means” prediction that those cells will become active the next timestep, or it just shares the same chemical depolarization effect?
This statement doesn’t make sense to me. The distal input does not represent prediction, or mean prediction. These signals always represent active neurons within a range of current synapses. They can cause neurons to become predictive.