Evaluating Fine-Tuning Efficiency of Human-Inspired Learning Strategies in Medical Question Answering

TL;DR

This paper assesses human-inspired data ordering strategies for fine-tuning large language models in medical question answering, highlighting their variable effectiveness and the promise of model-generated difficulty labels for cost-efficient training.

Contribution

It systematically evaluates five human-inspired strategies across multiple models and datasets, revealing their variable success and introducing LLM-generated difficulty labels as a promising alternative.

Findings

Interleaved strategies achieve the highest average accuracy gains.

Best strategy effectiveness varies across model-dataset pairs.

LLM-defined difficulty labels outperform human labels in curriculum learning.

Abstract

Fine-tuning Large Language Models (LLMs) incurs considerable training costs, driving the need for data-efficient training with optimised data ordering. Human-inspired strategies offer a solution by organising data based on human learning practices. This study evaluates the fine-tuning efficiency of five human-inspired strategies across four language models, three datasets, and both human- and LLM-labelled data in the context of medical question answering. These strategies achieve the best accuracy gain of 1.81% and an average gain of 1.02% across datasets, with interleaved strategies delivering the best average results. However, the best strategy varies across model-dataset combinations, limiting the generalisability of the effects of any single strategy. Additionally, LLM-defined question difficulty outperforms human-defined labels in curriculum-based learning, showing the potential of model-generated data as a cost-effective alternative for optimising fine-tuning.