Polychronous Wavefront Computation (PWC) defines a simple model of spiking neural networks based on spatially positioned transponders stimulated by signal wavefronts. The simplicity of the concept and elimination of explicit connections between transponders can potentially provide a practical basis for construction of large scale, complex pattern recognition systems. Previous work has shown how PWC transponders can be used to perform basic computations and be organized into simple pattern recognition configurations but the creation of complex pattern recognition behavior remains a difficult problem. The purpose of this work is to identify the key characteristics of complex biological and artificial neural networks and explore the application of those characteristics to the PWC model. The analysis includes a review of neuromorphic processes such as spike-timing-dependent plasticity, synaptic fatigue and potentiation decay as well as biologically inspired artificial neural network structures such as multi-layer perceptrons and convolutional neural networks. Recognition of stimulus sources and use of inhibitive stimulus are identified as key characteristics not addressed by PWC. Approaches to incorporate these key characteristics into the PWC model are discussed with the objective of creating transponder configurations that use unsupervised learning to perform complex pattern recognition.
Read full abstract