SOIM - A SELF-ORGANIZING INVERTIBLE MAP WITH APPLICATIONS IN ACTIVE VISION

We propose a novel neural network called the self-organized invertible map (SOIM) that is capable of learning many-to-one functionals mappin gs in a self-organized and on-line fashion, The design and performance of the SOIM are highlighted by learning a many-to-one functional mapp ing that exists in active vision for spatial representation of three-d imensional point targets. The learned spatial representation is invari ant to changing camera configurations, The SOIM also possesses an inve rtible property that can be exploited for active vision, An efficient and experimentally feasible method was devised for learning this repre sentation on a real active vision system, The proof of convergence dur ing learning as well as conditions for invariance of the learned spati al representation are derived and then experimentally verified using t he active vision system, We also demonstrate various active vision app lications that benefit from the properties of the mapping learned by S OIM.