Improved self-organization with an alternative inhibition gradient in CALMMap networks, Con.pdf

1Improved Self-Organization with an Alternative Inhibition Gradient in CALMMap Networks Adriaan Tijsseling Abstract Phaf, Den Dulk, Tijsseling Lebert (2001) described a self-organizing variant of the CALM modular neural network model (Murre, Phaf, Wolters, 1992), which has been coined CALMMap. The self- organization mechanism is driven by a specific inhibition gradient formula. The disadvantage of this formula is that it introduces new parameters and is not optimally adjusted to the size of a module. This research note provides an alternative for this inhibition gradient formula with the benefits that it performs autonomously. In other words, no extra parameters are introduced and the strength of inhibition adapts to the size of a module. Simulations are described to show the effectiveness of the alternative inhibition gradient function. 1 Introduction CALMMap has been introduced by Phaf, Den Dulk, Tijsseling Lebert (2001) as a self-organizing extension to the Categorization-And-Learning-Module network by Murre, Phaf Wolters (1992). CALMMap shows some practical improvements over both the classic version of the self-organizing feature map of Kohonen (1988) and the standard CALM network. CALMMap needs no external regulating mechanisms, but is capable of dynamically adjusting its learning rate on the basis of the degree of detected novelty, and shrinks its “activity-bubble” without adjusting the inhibitory weights (Phaf et al., 2001). Activations in CALMMap are also determined by a weighted-sum rule instead of by an Euclidean distance measure, allowing for improved separation between representations. A drawback with CALMMap is that the function that is used to create the inhibition gradient of the intramodular weights from veto-nodes to representation-nodes (for details see Section 2) introduces three extra parameters that needs to be fine-tuned for each module. In this report, we provide an alternative that does not introduce extra parameters and also generates an