Improvised Progressive Neural Network (iPNN) for Handling Catastrophic Forgetting

Learning is a technique which is used to solve a complex sequence of tasks like catastrophic forgetting. Transfer of learning starts from a pre-trained model in the source domain and learning gets transferred to a target domain. A progressive neural network is a technique in the eld of transfer learning, to solve the problem of catastrophic forgetting. Catastrophic forgetting is an area of machine learning, in which machines forget the previously known data while training a new model in transfer learning. The contribution of this paper is a novel technique called improvised Progressive Neural Network (iPNN) to handle the problem of catastrophic forgetting. The proposed technique reduces the loss of information while transferring knowledge from one model to another model in transfer learning. In this technique, an improvised progressive neural network is presented such that the storage constraints are significantly reduced The basic idea of the technique is to remove those models whose knowledge is no longer required in other models for a long period It also removes that model whose knowledge is already transferred in another model, completely or partially. In this case, the target model already captures the complete knowledge of the source model. With the removal of such source models, performance in terms of computation costs is significantly enhanced