Jevi Waugh

2026 Research Experiences Showcase

Fri, 27 Mar 2026 09:00:00 +0000

I have recently been selected as a presenter for the RP Showcase based on my recent summer research program.

The following information is from the faculty:

The Student Enrichment and Success unit warmly invites you to the 2026 Research Experiences Showcase!

Join us in celebrating the achievements of students from the 2026 Summer and 2025 Winter Research Programs. Enjoy inspiring student presentations that highlight their research discoveries and personal journeys, while engaging with the research community, building meaningful connections, and discovering how this program supports students in shaping their futures.

Event Agenda: 9:00am - Welcome 9:15am - 11:00am - Presentations 11:00am - 11:30am - Networking & Morning Tea

Event Registration opens at 8:30am with the formalities commencing at 9:00am. Attendees must be registered and seated by 9:00am.

Note that you will have to book in through the student hub portal either as a student/alumni or staff.

2026 Research Experiences Showcase Talk

Fri, 27 Mar 2026 09:00:00 +0000

Watch the Talk Recording

Learning frequency domain codes for semantic vision

Fri, 27 Mar 2026 09:00:00 +0000

I am a contributing author on the research paper Learning Frequency Domain Codes for Semantic Vision, currently under revision for resubmission.

Analysis of Gallstones

Fri, 01 Aug 2025 00:00:00 +0000

Machine Learning-Based Classification of Table Tennis Swings Using Racket Kinematics

Wed, 07 May 2025 00:00:00 +0000

This work is driven by the results in my previous paper on LLMs.

Create your slides in Markdown - click the Slides button to check out the example.

Add the publication’s full text or supplementary notes here. You can use rich formatting such as including code, math, and images.

Leveraging Persistent Homology for Topological Feature Extraction in Machine Learning: The PH-AML Pipeline

Tue, 01 Apr 2025 00:00:00 +0000

Click the Cite button above to demo the feature to enable visitors to import publication metadata into their reference management software.

Create your slides in Markdown - click the Slides button to check out the example.

Add the publication’s full text or supplementary notes here. You can use rich formatting such as including code, math, and images.

Fine-Tuning FLAN-T5 for Biomedical Lay Summarisation (BioLaySumm 2025)

Wed, 01 Jan 2025 00:00:00 +0000

This page provides a high-level overview only.
For full methodology, experiments, training scripts, and results, see the GitHub repository.

Overview

This project focuses on biomedical lay summarisation, translating expert-level radiology reports into language accessible to non-experts. The work fine-tunes instruction-tuned FLAN-T5 models on the BioLaySumm 2025 dataset and systematically evaluates different adaptation strategies.

Methods

Three optimisation strategies are explored:

Full Fine-Tuning (FFT): updates all model parameters.
LoRA (PEFT): updates a small set of low-rank adapter weights (~2–3% of parameters).
Evolution Strategies (ES): gradient-free optimisation via population-based parameter perturbations.

Performance is evaluated using ROUGE-1/2/L/Lsum, with analysis of compute cost, convergence behaviour, and parameter efficiency.

Key Findings

LoRA achieves comparable or slightly higher ROUGE scores than full fine-tuning while training only ~2–3% of parameters.
FLAN-T5-Base + LoRA provides the best balance of quality and efficiency.
Evolution Strategies offer fast iteration but underperform gradient-based methods for this task.

Example Visuals

Reproducibility & Code

All experiments were run on NVIDIA A100 GPUs using PyTorch and Hugging Face Transformers. Training scripts, hyperparameters, datasets, and seeds are fully documented in the repository.

👉 Full code and documentation:
https://github.com/Jevi-Waugh/BioLaySumm-Flan-T5/tree/topic-recognition/recognition/FLAN-T5-Jevi-Waugh