NeuroBatch: Harmonising data from an international neuroscience project
Project Overview
As part of this internship, the group have asked me to investigate whether or not a small-scale Large Language Model (LLM) has the ability to assist in data curation. More specifically, we are looking at the translation of human-submitted resting state EEG metadata into a machine-readable format, such as a JSON file. This translation is a key step in a pipeline designed to convert resting-state EEG datasets into BIDS format, with the overarching aim to use this pipeline to harmonise diverse, researcher-donated data for so that it can be integrated into a large, centralised database. However, a difficulty in this pipeline is that portions of the metadata may be provided in unstructured natural language, and at the moment, this information must be manually extracted to ensure it meets the requirements of the processing pipeline. This project aims to see if a small-scale LLM can be used to reliably automate this process, and structure it in a machine readable template for use in the pipeline.
What are the key results of your project?
- LLM was successful at parsing dataset information from large chunks of natural language
- LLM was able to parse information from a spreadsheet
- LLM was able to combine the above points and produce a JSON format so that extracted information was machine readable
- The above processes could be completed in around 10-15 seconds compared to a human taking around 5-10 minutes (if experienced in BIDS)
How do you feel you have benefitted from completing this internship?
This Project has been highly beneficial to me and my learning throughout this academic year. Firstly, it has allowed me to gain invaluable knowledge of the research process within psychology, and I have had the opportunity to work with a team of postgraduate researchers, broadening my knowledge and improving my academic and research skills while meeting some wonderful people.
The project has also caused me to improve my time management to a great degree in order to be able to balance my academic work, my personal life and the responsibilities involved within the project. It has caused me to challenge my personal perspectives and biases, as I was incredibly sceptical and hesitant about the use of AI before starting this project. Some of my concerns did seem to be valid; however, many of them were also proven false, and I have become more accepting and understanding of how AI and LLMs can be used within research and various industries.
Overall, this project has been an amazing experience, allowing me to build upon a variety of skills and join a network of incredible academics whom I cannot thank enough.