Multi-level neural networks

One of the challenges the neural network faces today is that the network lacks memory capability. Typically, it is a very time-consuming process to train a neural network for a given problem. However, when a new problem is presented, the neural network has to go through the learning process again without any benefit from the previous learning process. In order to address this problem, we propose a new architecture of neural networks that have the capability to remember past experiences and utilize them to solve a new problem faster and more efficiently. The proposed neural network consists of a primary neural network and a number of control neural networks. Preliminary experiments provide some promising results.