In SOMMA, each modal data flow is processed in a generic cortical map. On the figure, the map defined as visual, because she receives visual data, is structurally identical to the auditory map, that receives auditory data.
Each map is composed of a bidimensional set of generic units, inspired by cortical columns.
Each column has three layers, each layer has semantically distinct function and incoming data flows.
The sensory layer receives modal data and provides a first representation of the current stimulus. This representation is based on learning a tabular coding to a monomodal correlation, i.e. that this layer activity depends on a similarity function between a correlation, said discriminated, and the current stimulus. At the map level, discriminations of the sensory layer are self-organized meaning that two close columns have close discriminations.
The cortical layer provides partial information on perceptions made in other modalities. Perception of a modal correlation in a map activates the cortical layers of columns in other maps representing correlations that are multimodally linked to the perceived correlation. This layer is also implicated in learning of multimodal relationship between monomodal correlations.
The perceptive layer integrates modal and multimodal data to provide an unified multimodal perception of the current stimulus. This perception is a continuous spatial coding of the current stimulus that uses the sensory layers self-organization.