An artificial or a biological agent catches its internal state as well as the state of its environment thanks to numerous sensors. These sensors provide data from several senses (proprioception, vision, audition, ...). In each sense, multiple modal information are available (for example colour, speed or shape for vision). In my work, unification of multiple modal artificial data is inspired by psychological experiments and by sensorimotor theories. This unification is based on detecting correlations of the current multimodal stimulus. A correlation is a temporally recurrent spatial pattern that appears in an input flow. A correlation is defined as monomodal (respectively multimodal) if the pattern is included in one (respectively several) modality(ies).
This thesis proposed some functional paradigms for multimodal data processing, leading to the connectionist, generic, modular and cortically inspired architecture SOMMA (Self-Organizing Maps for Multimodal Association). In this model, each modal stimulus is processed in a cortical map. Interconnection of these maps provides an unifying multimodal data processing. Learning of input flow correlations space consists on sampling this space and generalizing these samples. These sampling and generalization of correlations are based on the constrained self-organization of each map.
The model is characterised by a gradual emergence of these functional properties: monomodal properties lead to the emergence of multimodal ones and learning of correlations in each map precedes self-organization of these maps. Furthermore, the use of a connectionist architecture and of on-line and unsupervised learning provides plasticity and robustness properties to the data processing in SOMMA. Classical artificial intelligence models usually miss such properties.