ABSTRACT
This chapter considers a particular class of artificial neural networks (ANN), known as reservoir computing networks, which combine aspects of recognition and decision-making required by regenerative processes observed in planaria. It analyzes reservoir network training methods that improve generalization to noisy input and robustness to network perturbation and explore an ANN experiment similar to the phenotypic reprogramming that occurs in regenerating planaria. The chapter illustrates the concepts by analogy to planarian regeneration, this work is meant to be a proof-of-principle example that can potentially apply to many biological contexts. It explores modifications to include noisy input and allow reservoir rewiring during training and evaluate its performance under novel noisy input as well as perturbation. The chapter demonstrates that a reservoir network model can learn a pattern of planarian regeneration and emphasize parallel results to planarian patterning under gap-junction perturbation.
