Shape representation and learning process.

在这个项目中，我们试图从数据压缩和恢复的角度来解释心理相似性和困惑矩阵(Inui等人，1996)。接下来，我们考察了心理相似性和困惑矩阵如何依赖于呈现给受试者的刺激集(Makioka等人，1995)。此外，我们还考察了这些变化是如何通过监督分类区别学习(Misaki，1998)来实现的。我们的实验结果可以通过对刺激模式的针尖成分分析来近似解释。我们的神经网络模型可以很好地解释辨别任务的学习曲线和心理相似性的变化。对于三维物体识别，Nishina和Inui(1996)发现存在两个独立的系统，即二维比较过程(如GRBF)和三维比较过程。

In this project, we tried to explain psychological similarity and confusion matrix from a view point of data compression and restoration (Inui et al., 1996). Next, we examined how the psychological similarity and confusion matrix depend upon the stimulus set presented to subjects (Makioka et al., 1995). Furthermore, we examined how these changes with supervised leaming of categorical discrimination (Misaki, 1998). The results of our experiments can be approximately cxplained by pincipal component analysis of the stimulus patterns. It appears that learning curves and changes of psychological similarity during discrimination task can be explained by our neural network model. For 3D object recognition, Nishina and Inui (1996) found that there exist two separate systems which are 2dimensional comparison process such as GRBF and 3 dimensional comparison process.