An integrated model of visual recognition with contextual dependency

Connectionism and symbolism have contrasting characteristics. This has motivated studies on the integration of these to achieve flexible and effective information processing. Major issues in integrated models are the representation of pattern and symbolic information, and the interaction between them. In the present paper, psychological experiments on the bowl/cup problem are carried out to clarify the effect of context on visual recognition. To explain the stochastic behavior of a subject during experiments, parameters representing cognitive boundaries of a bowl and a cup in each context are estimated by maximum likelihood based on one´s responses. An integrated model composed of a connectionist module, a symbolic module and a context module is proposed. Its parameters are estimated using data on one´s responses. The resulting integrated model is evaluated using data on one´s response times. This partition of data, i.e., responses and response times, constitutes a general framework for the modeling of psychological phenomena