UESMANN: A Feed-Forward Network Capable of Learning Multiple Functions

A number of types of neural network have been shown to be useful for a wide range of tasks, and can be “trained” in a large number of ways. This paper considers how it might be possible to train and run neural networks to respond in different ways under different prevailing circumstances, achieving smooth transitions between multiple learned behaviours in a single network. This type of behaviour has been shown to be useful in a range of applications, such as maintenance of homeostasis. We introduce a novel technique for training multilayer perceptrons which improves on the transitional behaviour of many existing methods, and permits explicit training of multiple behaviours in a single network using gradient descent.