Numenta Logo

Spatial Pooling

Spatial Pooling

The neocortex is composed of a hierarchy of regions. The feedforward input to each region consists of millions of nerve axons which come from sensory organs and other regions. Most regions in the neocortex receive input from multiple sources; it is not uncommon for a region to receive input from six or more other regions. These inputs are generally mixed together, so a region doesn’t “know” where the inputs originated or what they represent. Confusing the picture, the number of input axons doesn’t correlate closely with the size of the region. How can a region process inputs from many different sources without any prior knowledge of what these inputs represent, how many input bits there will be, and what spatial patterns may exist in the input? Spatial Pooling, a learning mechanism fundamental to both the neocortex and Hierarchical Temporal Memory (HTM), is the answer to this problem. While we are working on a standalone chapter for BAMI that will detail the Spatial Pooling algorithm, for now you can explore the following resources to learn more: