Training Compact Models for Low Resource Entity Tagging using Pre-trained Language Models
share
Training models on low-resource named entity recognition tasks has been shown to be a challenge, especially in industrial applications where deploying updated models a continuous effort and crucial for business operations. Often in such cases, there is abundance of unlabeled data, however, labeled data is scarce or unavailable. Pre-trained language models trained to extract contextual features from text were shown to improve many natural language processing (NLP) tasks, including scarcely labeled tasks, by leveraging on transfer learning. However, such models impose a heavy memory and computational burden, making it a challenge to train and deploy such model for inference use. In this work-in-progress we combined the effectiveness of transfer learning provided by pre-trained masked language models and use a semi-supervised approach to train a fast and compact model using labeled and unlabeled examples. Preliminary evaluations show that the compact models achieve competitive accuracy compared to a state-of-art pre-trained language models with up to 36x compression rate and run significantly faster in inference, thus, allowing deployment of such models in production environments or on edge devices.
READ FULL TEXT