Doctoral Assistant, University of Waikato.

Hamilton, 2018 - 2019

Teaching and marking duties for undergraduate computer science courses.

Financial and Systems Manager, Community Support Medical Centre.

Dunedin, 2011 – 2018.

Quality improvement, financial systems and tax compliance for a busy medical practice.

• Developed and implemented a remote access scheme for overseeing financial and IT security. This included implementing risk management against a world-wide cyberattack.
• Meeting project deadlines with minimal supervision (working remotely from Hamilton).
• Collaborating effectively with internal stakeholders (health professionals, practice manager) and external contractors.
• Handling sensitive financial and health information in a cyber secure manner.
• Co-ordinated a practice wide quality improvement project that achieved Foundation Standard accreditation from the New Zealand College of General Practitioners.

Lecturer, Manukau Institute of Technology.

Auckland, 2007 – 2010.

Lecturer for Mathematics and Computer Skills, School of Electrical Engineering and Trades.

• Chosen as the liaison between student reps and the faculty, and also as the International Student Advisor for the School. (Based on student review).
• Development of teaching material for adult learners.
• Researched and applied various techniques of incorporating technology to teach mathematics at different levels (Certificate, Diploma and Degree) and for different age groups.

Master of Science (Computer Science)

Automatic De-identification of Electronic Health Records using Word Embeddings.

Abstract: A study of de-identification of electronic health records using word embeddings and machine learning classifiers is presented. In the medical domain, de-identification has been proven to be an essential but challenging problem. Recent interests and developments are mainly driven by the de-identification competitions: the 2014 i2b2/UTHealth shared task and the 2016 CEGS N-GRID shared task. This study focuses on two aims. The first aim is to conduct a survey of the recent de-identification competitions. This includes the understanding of the datasets, the systems developed for these competitions, and any shortcomings and avenues for future research. The second aim of this study is to conduct a feasibility study on using word embeddings and machine learning classifiers to develop a well-performing, simple, and adoptable de-identification system. The 2014 de-identification competitions' dataset is used for the second aim. Word embeddings applying to the de-identification problem in the medical domain is an entirely novel approach. For this study, word embeddings are formed using the fastText model. Sliding window size variations to capture fastText output are explored. The machine learning classifiers used in this study include random forests, logistic regression, J48 and naive Bayes. WEKA was used as an analysis platform to perform all experiments with a focus on optimising the F-measure. Observed results from this feasibility study were similar to that of the 2014 competitions. Overall, the results obtained from this study are very positive and encouraging and open the door for future research. This study provides a great start to research and development of machine learning based, a well-performing de-identification system for medical text.

Predicting Transfer of Brain Trauma Patients using Machine Learning Techniques.

Abstract: A study of predicting transfers of brain trauma patients using machine learning techniques is presented. Patient transfers occur between tertiary, secondary and rural hospitals within the Waikato District Health Board (WDHB), as well as across New Zealand. The study dataset was provided by Midland Trauma Systems based at the WDHB in Hamilton city. The main focus of this study is the prediction of first transfer (if any) of patients based on clinical diagnosis data. A secondary study was also conducted on predicting factors that influence the time it took for a transfer to occur. The studies are treated as text classification and regression problems respectively. Machine learning algorithms used include: J48, JRIP, Random Forest, M5' and libLINEAR. WEKA was used as an analysis platform to perform all experiments. In all experiments, predictive error was high, suggesting that the available data is insufficient for predicting first transfer and the time it took for a transfer to occur.

Master of Science (Applied Mathematics) Thesis

The Phase error of Explicit Runge-Kutta-Nystrom Methods.

Abstract: The aim of this thesis is to investigate the phase error of explicit Runge- Kutta-Nystrom (ERKN) methods when used to perform long simulations of the Solar System. Since the detection of close approaches to planets by small bodies is important in many simulations, we are particularly interested in measuring the phase error with respect to the radii of the close approach spheres. We use the two- body problem and a model of the outer Solar System as test problems and investigate the phase error for low and high order symplectic and non-symplectic ERKN methods. We find for both symplectic and non-symplectic methods that higher order methods provided a marked advantage over low order methods.

Graduate Diploma in Tertiary Teaching

The use of computer technology in tertiary mathematics teaching.

Summary: Exploring options to integrate technology in tertiary mathematics teaching for students with diverse educational, age and ethnic backgrounds. The target student population was Certificate and Diploma students in the electrical trades. This included delivery through blended learning, e-learning and mathematical software. Practical implementation of the research findings in my workplace led to being selected as a finalist of Teaching Excellence Awards.

University of Auckland summer projects

Solar system modelling - Modelling horse-shoe orbit of an asteroid (which at that time was feared to be on a collision course with Earth).

Mathematics education - Literature review of studies on women in technology.