CATEGORIZATION AUTOMATA EMPLOYING NEURONAL GROUP SELECTION WITH REENTRY

摘要

An apparatus capable of sensing the presence of objects in its environment, categorizing these objects without a prior description of the categories to be expected, and controlling robotic effector mechanisms to respond differentially to such objects according to their categories. Such responses include sorting objects, rejecting objects of certain types, and detecting novel or deviant objects. The invention includes a device called a "classification n-tuple" (of which a "classification couple" is a special case) capable of combining signals from two or more sensory modalities to arrive at the classification of the object.PPThe invention operates by simulating certain features of animal nervous system, including neurons with responses determined at each moment of time by inputs received along synaptic connections from sensory means or form other neurons. In this invention, neurons are arranged in "groups", which permits them to act cooperatively while still retaining characteristic individual responses. Groups in turn are arranged in "repertoires", which provide a totality of response specificities sufficient to respond to any of a range of possible input objects. Some of these repertoires are organized as "maps", such that groups responding to inputs that are similar along some dimensional (spatial or abstract) are close together in the repertoire, enabling the repertoire to respond correctly to novel objects that are similar to objects it has encountered previously. Maps in this invention are linked by connections, called "reentrant" connections, that enable the responses of one map to be reentered into the system as inputs to another map. Reentrant connections permit signals from different sensory modalities to be correlated in an ongoing fashion, and they are the basis for the operation of classification n-tuples.PPThe responses of the apparatus of this invention may be improved and optimized for a particular task through the operation of "neuronal group selection", a process which enhances the responses of those neuronal groups in the apparatus which were active in some time interval preceding the production of a useful output action. This process operates by modifying the strength of selected synaptic connections. The value of various output actions is determined, for the purpose of regulating selection, by means which are internal to the apparatus of this invention. Accordingly, the invention is capable of self-organization and learning, and does not require a predetermined specification of correct responses for all possible inputs.