Developmental Artificial Neural Networks that integrates Functions using Grammar Variational Autoencoder

Abstract

In conventional deep learning, when a network topology obtained by training for a specific task is transferred to a different task, it is based on a single learned model that is expected to best fit the new task, as in fine tuning or transfer learning.On the other hand, organisms can abstract multiple previously acquired experiences and adapt them to new tasks as knowledge.Based on this, this study will use the framework of Weighted Agnostic Neural Networks (WANN) and Grammar Vatiational Auto Encoder (GVAE) to develop a Developmental Artificial Neural Networks (DANNs) capable of simultaneously transferring multiple functions that generates a neural network with new functions that integrate multiple functions.