Federated learning for natural language processing in medical domain

• Main Author: Saraf, Ishita
• Other Authors: Anupam Chattopadhyay
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Federated learning; Natural language processing; Chatbot; Medical; Conversational agent
• Online Access: https://hdl.handle.net/10356/175336
• Institution: Nanyang Technological University

Recent years have witnessed an unprecedented surge in interest and innovation in the field of Natural Language Processing (NLP), largely due to ground-breaking developments such as the creation of ChatGPT and other Large Language Model (LLM) applications. In today's landscape, data-driven deep learning algorithms have become the norm in the field of NLP. One of the biggest challenges faced by such data-dependent NLP applications is related to data scarcity and privacy. Federated Learning (FL) is a convincing solution to overcome the issues regarding confidentiality of training data and constraints imposed by inadequate data in specific domains. The main objective of this study is to build a medical assistance chatbot using federated learning to overcome the limitations posed by the private nature data in the medical domain. The study compares the performance of centralized training and federated learning for fine-tuning a BERT-based conversational agent on a medical conversation dataset, serving as a stepping-stone for future research into federated learning for LLM-led NLP applications.