Many AI researchers have claimed that perception and thought are mediated through large scale, compositional, competitive knowledge structures known variously as frames (Minsky), scripts (Schank), or schemata (Rumelhart). Traditional AI systems hand-engineer these structures in service of a particular task. This project seeks to show how a robot could learn such structures from raw sensor data, through interaction with its environment. The system uses a combination of self-organizing feature maps (SOMs), and a growing auto-associative memory to construct schemas, bottom-up, from raw sensor data.
The Constructivist Learning Architecture (CLA) is a model of infant cognitive development. This model is based on a constructivist information-processing approach to cognitive development, which postulates that Piagetic stages of development are a characteristic of infants' learning capabilities. The claim is that a single information processing mechanism can account for developmental change throughout development and across domain. CLA is implemented using a hierarchy of Self-Organizing Maps, in which higher-level maps learn patterns of activation in lower-level maps.
VISOR is a large connectionist system that shows how visual schemas can be learned, represented, and used through mechanisms natural to neural networks. Processing in VISOR is based on cooperation, competition, and parallel bottom-up and top-down activation of schema representations. Simulations show that VISOR is robust against noise and variations in the inputs and parameters. It can indicate the confidence of its analysis, pay attention to important minor differences, and use context to recognize ambiguous objects. Experiments also suggest that the representation and learning are stable, and its behavior is consistent with human processes such as priming, perceptual reversal, and circular reaction in learning. The schema mechanisms of VISOR can serve as a starting point for building robust high-level vision systems, and perhaps for schema-based motor control and natural language processing systems as well.