Sydney, Australia
I am a software engineer in industry, and this is my research website. I work in multimodal machine learning, and my aim is to develop models which learn the dynamics of the systems they observe from limited resources.
I was fortunate to be advised by Prof. Fred Roosta during my honours year at the University of Queensland, where I graduated with First Class Honours. I spent a wonderful summer in Prof. Patrick Slade's lab at Harvard University. I am also grateful to have worked with Dr. Jonathan Kummerfeld on clinical language models at the University of Sydney, and with Prof. Matthew Luskin at the University of Queensland. I currently collaborate with Prof. Piotr Koniusz at UNSW.
Introduces the adjoint operator of a nonlinear map to generalize importance sampling and its guarantees from linear models to neural networks.
Learns the latency between sensor modalities jointly with their representations, without timing labels, recovering the physical order in which events reach the sensors.
Adapts a global vision model to under-represented regions with limited labels, outperforming locally trained models most on out-of-distribution camera sites.
An open-source, nationally adopted workflow for training and deploying custom wildlife-image classifiers.
Inter-sensor latency learning, coreset guarantees for multimodal data, and hyperbolic representations with Prof. Piotr Koniusz (UNSW).
Multimodal clinical language models. With Dr. Jonathan Kummerfeld and Dr. Ben Hachey.
Adapting global vision models to local data, and the MEWC workflow. With Prof. Matthew Luskin and Dr. Dan Morris (Google Research).
Multi-sensor models for hip-exoskeleton assistance, deployed for real-time inference on an embedded GPU. With Prof. Patrick Slade.
Importance sampling for nonlinear models, the basis of the ICML paper. With Prof. Fred Roosta.