Stable sparse encoding for predictive processing

Hierarchical Predictive Coding systems that have adopted prediction as their primary goal, are heavily reliant on the stable sparse coding of sensory input. Furthermore, such systems will require their spatial coding function to be adaptive and able to reform to reflect changes within the environment. These properties of stability and adaptiveness should emerge naturally from the spatial coding system and not be reliant on additional control mechanisms. Hierarchical Temporal Memory is a cortically inspired model that encapsulates both sparse coding and temporal processing functions. We present an investigation into the stability and adaptiveness of three alternative versions of the spatial pooling function. Our results show that two of these SP algorithms are able to form stable sparse distributed representations of audio input, while still remaining adaptive to changes within the input data.