Modalities merging in Somma has the following functional properties:
- sampling: learning of some input flow correlations
- generalizing: generalizing learned samples to input flow correlation space
- unifying: merging of modal stimuli by searching a present multimodal correlation in the stimuli
Cortically inspired properties of Somma are:
- separated processing of modal data to reduce learning space dimension
- local and decentralized learning and computation (connectionism) leading to lesion robustness
- unsupervised learning leading to autonomy of the model
- on line learning (no distinct learning and exploitation phases) and generic architecture leading to plasticity of the system to changing inputs