R-trees vs arrays/hash-maps vs combined methods for calculating local inhibition

[breznak]

@SimLeek

nd the complexity of finding those points is O(log_m(n)),

We have also problems with the speed of local inhibition, and quality of global.
@dmac proposed improvement to SpatialPooler (includes macrololumns within each macrocol global (fast) inhibition happens, communication outside the macrocol is considered too far (there is none))

breznak

This helps combine semantic quality of local inh with speed of global.
Actially helps achieve even better results in our MNIST test.

@SimLeek

Hashmaps could also be used to store x and y locations with O(1) query tim

I would be interested in this implementation and discussion of its problems, as I propose replacing inhibition (resp. the neighbourhood search) with proximal connections

Replace inhibition with proximal horizontal inhibitory connections

The mechanism for Inhibition is well known and accepted. The implementation uses (topological) neighbourhood in which cells compete for activation (k-winner-takes-all). Now, the problem is with biologically defining the neighbourhood. I propose to add proximal horizontal (to peer cells) connections within Spatial pooler. peer connections randomly sample from distribution, same as feed-forward conns. a “negative” Hebbian rule applies: each time 2 cells fire (activate) together, their peer co…