Projects

Restless Multi-Armed Bandits

Supervisor: Yoni Nazarathy

I am working on a project which aims to provide theoretical and empirical insight into a restless multi-armed bandit problem. We are exploring a range of models in both discrete and continuous time, and are interested in finding suitable decision policies.

In particular, we are studying the interaction between exploration and exploitation in partially-observable models. We are also interested in simulation with Julia and the analysis of numerical results.

Reliability Estimation in Dynamic Networks

Supervisor: Thomas Taimre

I completed a semester-long research project which aimed to modify rare-event Monte Carlo estimators of static network reliability and adapt them for estimating the reliability of dynamic networks.

I created a suite of simulators implementing a number of rare-event simulation techniques, and tested their effectiveness for several networks under different models and assumptions. I also presented my research results in a talk, and received a high distinction for my final report.

Brasileiro

Brasileiro is a full-stack sheet music archive for browsing, searching, and opening thousands of PDF music scores from a central collection. The project includes a statistical learning PDF processing engine which at its core uses a random forest classification model for processing large PDF songbooks.

The front-end is in React and Typescript, back-end with Django DRF. Deployed using B2, Cloudflare, and a VM.

For copyright reasons, the PDFs are not accessible publically but the codebase is open-source.

AusStats.jl

AusStats.jl is a Julia package for finding, downloading, reading, and tidying Australian Bureau of Statistics data.

It is designed for DataFrame-first workflows around ABS data, including discovering publications, caching downloads, parsing time-series workbooks into tidy observations, reading data cubes, and querying ABS API endpoints.