Publications by authors named "Dan Schofield"
Artif Intell Med
December 2024
Article Synopsis
- - Pre-trained Large Language Models (LLMs) show promise in Natural Language Processing (NLP) but face challenges when applied to specialized fields like healthcare due to resource constraints and specific language requirements.
- - The study investigates efficient methods to adapt smaller LLMs for healthcare tasks while addressing limited computational resources, focusing on three pre-training approaches: Masked Language Modeling, Deep Contrastive Learning, and a novel metadata-based method.
- - Results indicate that contrastively trained models surpass other methods in classification accuracy, even with fewer labeled data, underscoring the significance of domain specialization, though the metadata approach did not enhance classification performance further.
