Joint learn: A python package for task-specific weight sharing for sequence classification

Abstract

Transfer Learning has enabled cutting-edge improvements in Deep Learning to attain state-of-the-art outcomes, notably in the domain of Natural Language Processing. Despite this, neural networks trained on the low-resource text classification corpora still face challenges because of the lack of pre-trained model checkpoints. In this paper, we introduce Joint Learn which is a PyTorch based comprehensive toolkit for weight sharing for text classification that leverages task-specific weight sharing to train a joint neural network for several sequence classification tasks and aids in the development of more generalized models while potentially eliminating the transfer learning issues that low-resource corpora encounter.