Konzeptionelle Modellierung geometrischer Invarianzen in der visuellen Wahrnehmung von Primaten - Situativ gesteuerte Complex-Bildung als Grundlage invarianter Zellantworten

Table of Contents: Our sense of vision conveys a stable percept of the environment. We can recognize objects regardless of the viewpoint we take. Available neural models can only partly explain this invariant perception. Standard models are based on hierarchic arrangements of nerve cells. They aim at constructing specific neural responses to complex visual stimuli from responses to simple stimulus components. A main concept is the neural realisation of logical OR operations by convergent projections (complex-formation). However, generation of invariance to certain stimulus variations is in conflict with the formation of stimulus specific responses -- both at the signal level, and as a model of thinking. Consequently, classical models of invariant visual shape perception show weaknesses, like the binding problem, and the inability to distinguish objects with overlapping sets of retinal projections. I present a conceptual modelling approach to the issue. Living beings actively explore their environment: External physical conditions do influence processing in the visual system. Here, I formulate the concept of situation controlled complex-formation, based on the fact that external parameters exert control on neural transfer properties. As an application of this concept, I present two models of invariant visual processing: (1) neural retinal slip correction, and (2) distance invariant representation of visual objects. My models overcome substantial problems of classical modelling, at the cost of increased neural resources. For the model of distance invariance, the concept of situation controlled complex-formation predicts a new cell class, the distance complex cells. Recent experiments report the finding of neurons with similar properties. In both models, situation controlled complex-formation generates scene representations, which are independent of the controlling parameter. Most probably, the same method can produce invariance to other external conditions, too. Situation controlled complex-formation hence presents a universal tool for conceptual modelling of neural invariances. Moreover, it is an effective model of thinking in our understanding of cortical processing.