Zero-shot and Few-shot Generation Strategies for Artificial Clinical Records

arXiv:2403.08664v1 Announce Type: cross
Abstract: The challenge of accessing historical patient data for clinical research, while adhering to privacy regulations, is a significant obstacle in medical science. An innovative approach to circumvent this issue involves utilising synthetic medical records that mirror real patient data without compromising individual privacy. The creation of these synthetic datasets, particularly without using actual patient data to train Large Language Models (LLMs), presents a novel solution as gaining access to sensitive patient information to train models …

abstract artificial arxiv challenge clinical clinical research cs.cl cs.lg data few-shot issue medical medical records patient privacy records regulations research science strategies synthetic type zero-shot