A Computational Model Based on Self-Organizing Synaptic Formation for Motion Direction Detection

The formation of direction-selective visual circuits is thought to involve the progressive refinement of synaptic connections during development. In biological visual systems, patterned spontaneous activity, such as retinal waves, has been proposed to provide structured spatiotemporal activity that contributes to the refinement of visual pathways before mature sensory experience is fully established. Motivated by this view of activity-dependent circuit organization, this study develops a Self-Organizing Map-Based Artificial Visual System, termed SOM-AVS, to examine how organized connectivity may emerge in a motion direction-detecting circuit. In the proposed model, local motion-detecting units extract elementary direction-related responses from visual input and project them to a global motion direction layer represented by a self-organizing map. Connections are progressively reshaped by winner selection and local cooperative updating, allowing initially unstructured connections to gradually acquire organized direction preference. After repeated exposure to generated retinal-wave-like activity data, the SOM layer develops topographically arranged regions corresponding to distinct motion directions. This organization suggests that direction-related response domains can emerge from activity-dependent self-organization without externally imposed labels. The proposed model should be regarded as a biologically motivated computational abstraction rather than a direct physiological reproduction of retinal-wave-driven circuit development. Within this scope, the model provides a computational framework for examining how retinal-wave-like activity and self-organizing plasticity may contribute to the formation of motion direction-related connectivity, offering a possible developmental interpretation for bio-inspired visual motion processing.