The fortnightly BRAG meeting will be held this Thursday 10th July at 1pm in GP-Y801 (and via Zoom). This week we will have presentations by Sishou Zhou and Mackenzie Parker.
Sishou’s Talk
Title: Fast and Functional, a physics-informed neural network (PINN) for parameter estimation in diffusion MRI
Abstract: Diffusion-weighted magnetic resonance imaging (DW-MRI), a specialised branch of MRI, allows for the non-invasive investigation of tissue microstructure by making signal measurements sensitive to the diffusion of water molecules. In biological tissues, this diffusion is often hindered and restricted, and can be effectively modelled using a time-fractional diffusion model, or sub-diffusion model. However, traditional parameter estimation methods for the sub-diffusion model, such as least squares, often perform poorly in the presence of noisy diffusion MRI data and can be very time-consuming. In recent years, physics-informed neural networks (PINNs) have emerged as a powerful approach in solving inverse problems. In this talk, we will present a novel PINN framework to enhance the accuracy and efficiency of parameter estimation in the sub-diffusion model for dMRI data, leading to more precise parameter maps of the brain.
Mackenzie’s Talk
Abstract: This research addresses a critical gap in competitive swimming by developing predictive, simulation-capable models for deriving personalized pacing profiles that explicitly capture uncertainty and heterogeneity in athlete performance. Leveraging functional data analysis, including Functional Principal Component Analysis (FPCA) with smooth spline bases, the study transforms high-resolution race data (e.g., from the SPARTA system) into functional representations and extracts key functional principal components (FPCs) that correspond to established pacing characteristics such as drop-off, back-end kick, end-spurt, and lap transitions. Clustering techniques – including K-means, hierarchical, and Gaussian mixture modelling (GMM) – are employed to try and develop unsupervised pacing profiles from the data. Validation is performed using the Calinski-Harabasz, Silhouette, and S_Dbw internal validation indices, ensuring that the selected number of clusters is justified by consistent trends across multiple metrics. The distribution of the data has produced complications in deriving validated and justified clusters, prompting further investigation into alternative profiling methods. Ultimately, when integrated with Swimming Australia’s analytical platforms, this research would not only advance theoretical understanding of pacing dynamics but also provide actionable insights for performance optimization to ultimately enhance strategic decision-making in elite swimming competitions.
Join Zoom Meeting
https://qut.zoom.us/j/87331850338?pwd=KajF5X7le4nJ9F2VgwgpCm8SS1bRJg.1&from=addon
Meeting ID: 873 3185 0338
Passcode: 926037