Reconstruction of Visual Sensory Space on the Hidden Layer in Layered Neural Networks

In layered neural networks, the input space is reconstructed on the hidden layer through the connection weights from the input layer to the hidden layer and the output function of each hidden neuron. The connection weights are modi ed by learning and realize the transformation to emphasize necessary information and to degenerate unnecessary one for calculating the output. In this paper, visual sensory signals are adopted as the input. In order to examine the reconstruction, (1)supervised or reinforcement learning is applied to a layered neural network at rst, (2)all the connection weights from the hidden layer to the output layer are reset to 0, (3)another supervised learning using some training data is applied, and nally (4)the output for the test data is compared to that when the rst learning was not applied. It is shown that the necessary information to generate the desired output in the rst learning was extracted on the hidden layer.