78 — Simple Cortex: A Model of Cells in the Sensory Nervous System

David Di Giorgio presents some very interesting concepts toward neuro-inspired computing — but before that, there are quite a few pages of very informed, high-level (sometimes slightly opinionated) writing about our modern understanding of sensory cortex. It covers the basics of neuron morphology and the hypothetical columnar motif structures of cortex: If you’re interested in neuroscience and don’t have a great foundation, I definitely recommend this paper.

Di Giorgio’s main technological contribution of this paper is Simple Cortex, a framework for machine intelligence that uses neuromimetic sensory cortex as a template and inspiration. It uses HTM — hierarchical Temporal Memory — developed by Numenta, to retain a degree of working knowledge of its inputs over time. This model is similar to the LSTM model currently gaining popularity in conventional machine-learning circles, but it operates on spiking neurons, rather than the artificial, continuous-output neurons of computational networks.

The paper includes a really exciting demonstration of SC “learning” the rules of physics in a 2D environment, where it correctly predicts the path of a ball after seeing only a few frames of its trajectory.

I’m excited to learn more about this neuro-inspired learning; it looks like the SC project is still very alive and in active development.