transfer learning based Neural networks

Transfer learning based based Neural networks is a research problems that works on pre-trained neural networks to extract low level features and the existing knowledge of the source ANN transferred  to a target ANN. Basically, four types of initialisation exists in transferred learning which are fully pre-trained ANN, partially pre-trained ANN, partially frozen ANN with random initialisation and partially frozen ANN with pre-trained initialisation. Fully pre-trained ANN – This method first considers a randomly initialised source ANN to pre-train the low level features which then transfers the existing weights and biases to the target ANN as one of the hidden layer. This method is suitable for structures with same configuration.