Published Paper Details:

: Most of the Deep Neural Network models like RNN and CNN are data hungry models. We may or may not have the huge dataset available to train those models. Thus, we require a model or an architecture which can use the pre-trained model to train the new model, which comparatively reduces the training time and gives good accuracy. The proposed model is a Convolutional Neural Networks (CNN) based model that is obtained by using few-shot transfer learning with a simple word embedding model to train the given DBPEDIA dataset. With a large DBPEDIA dataset available, the proposed CNN based model aims to extract the local and position-invariant features comparatively faster than the RNN based model. The few shot transfer learning aims to recognize concepts from the pre-trained word embedded models and extracts some common knowledge from the sequence of input data of training dataset. The few-shot transfer learning mechanism leverage the knowledge from the common source domains to special target domain with few data supporting the CNN model to extract the relevant feature efficiently. Thus, the proposed model, Few-Shot Transfer Learning based on CNN model, exhibits significant classification performance compared to other alternative models.