An approach to control aging rate of neural networks under adaptation to gradually changing context

Abstract

The paper presents a decayed prior sampling algorithm for integrating the existing knowledge of a supervised learning neural networks with the new training data. The algorithm allows the existing knowledge to age out in slow rate as a neural network is gradually retrained with consecutive sets of new samples, resembling the change of application locality under a consistent environment. The experiments are performed on 2-dimensional partitions problem and the results convincingly confirm the effectiveness of the technique.