sylvander.phd

Luke Sylvander, PhD

Computational scientist in Brisbane working on simulation, statistical inference and machine learning for noisy scientific data. PhD in Applied Physics from RMIT University.

The question I keep returning to is how to learn about a system when the model is expensive to run and the data only reach it indirectly. My PhD at RMIT (supervised by A/Prof. Jim Partridge and Prof. Dougal McCulloch) approached it from the sensing side: I built an electrolyte-gated biosensor for the motor neurone disease biomarker TDP-43, then trained spiking neural network classifiers to recover the biomarker from drifting, low-signal electrochemical measurements where threshold analysis failed. The work involved characterising noise, engineering features from raw device output and validating learned models against experimental data, and produced two first-author papers.

The simulation side came first. During Honours at QUT under Dr Konstantin Momot I wrote a molecular dynamics engine for a Lennard-Jones fluid from first principles and used it to estimate rotational diffusion constants across gas, liquid and supercritical phases, with implications for MRI contrast theory. Most of that project was spent on the practical statistics of an expensive stochastic simulator: sampling noise in correlation functions and finite-size effects.