Abstract: The perceptual boundary effects were investigated with connectionist model simulation. The relationship of perceptual boundary effects and knowledge structure was studied. The perceptual boundary effects of continuous XOR were studied with cascade correlation model of connectionism. 25 patterns which were distributed uniformly at the XOR plane were to be learned at the training stage. Three groups of other patterns were used at the test stage. There were 8 patterns in each group. These patterns had different distances to the boundaries. The test errors were recorded and compared for the different groups of patterns. For investigating the different knowledge structure, the contribution of each hidden node was recorded and studied by principal component analysis. The perceptual boundary effects of neural networks which represented different knowledge structure were compared. The results showed that the connectionist model could simulate the perceptual boundary effects, there was no significant difference between neural networks which had different knowledge structures for continuous XOR

Key words: connectionism, cascade correlation, perceptual boundary effects