A Hierarchical Brain Inspired Computing System
Christopher Johansson, Anders Lansner · 2006
In this paper we describe an abstract hierarchical model of the neocortex and its implementation in a connectionist network. This network has convergent feed-forward projections going from lower to higher levels in the processing hierarchy implementing competitive learning. There are also associative divergent backward projections going from higher to lower levels and associative recurrent projections at each level. Here we show that this type of multi-level attractor network extends the capabilities of a single layer network by enabling autoassociative storage of patterns not possible to store in the latter.