Self-organizing analog fields (SOAF)

Summary form only given. A neural network paradigm called self-organizing analog fields (SOAF) learns many-to-many, analog (as opposed to winner-take-all, WTA) spatiotemporal mappings. The learning mechanisms employed are unsupervised learning of features and self-organization of connectivity. Homeostasis is defined, providing a necessary extra degree of freedom to solve the credit assignment problem between these two different forms of learning. It is also shown that homeostasis provides a mathematical equivalent to the Lyapunov function for this paradigm. SOAF is hierarchical, allowing unlimited layers for spatiotemporal feature extraction/abstraction. Each SOAF layer is modular and mathematically self-contained, allowing arbitrary connectivity between layers. Empirical results indicate that unsupervised WTA is a degenerate case of SOAF; SOAF´s clustering performance is bell-shaped with respect to its primary tuning parameters; and peak clustering performance (top of bell) can be rapidly achieved and the peak broadened and heightened with more layers