Few grand challenges are more daunting than the intertwined mathematical, neurological, and cognitive mysteries of the brain. This monograph addresses all three, beginning with a simple new neural model that uses spike timing, approximates known neural behavior, and is capable of fast learning. The model is validated by both theory and simulations.
A new Shannon information metric (bits/neuron) that measures the recoverable learned information that resides in synaptic strengths is then derived. By optimizing the learning metric and its related functions, optimum training protocols and neural parameters emerge.
This neuron model is extended to neural layers and how they can be taught sequentially without pattern-specific training signals. A spike-based neural network dual-flow feedback architecture is then discussed as a likely key element of cognition, and is partially validated by natural experiments involving visual anomalies and mild hallucinations.
Models for Neural Spike Computation and Cognitionby David H. Staelin and Carl H. Staelin
Models for Neural Spike Computation and Cognition downloadable PDF