We think of the knowledge of task achievements in brain as the whole computation process involving neural network connection and their topology. Fast adaptation to a task by a knowledge reuse can be regarded as a survival value. We present a neural network model for fast adaptation to a task by reusing the knowledge derived from similar tasks. The model reuse the knowledge of the similar tasks. The knowledge are the structure and weight of the network connection. To validate the model, we experimented two computer simulations of navigational task with different but similar environments. In the first task A, a learning agent is required to move along the side walls. Next, in a task B, the agent required to move along the side walls and turn toward a light. We assumed that the agent had gotten the knowledge of task A, moving along the side walls. We compared the two case in the task B, a case that the agent reused the knowledge of the task A and the other case that the knowledge is not reused. We show that the navigation agent achieved the task B quite quickly reusing the knowledge of the task A.
展开▼
