Mr Jonas Harnau
I am an econometrician fascinated by deep learning.
My research as a DPhil candidate in Economics at Oxford gives me a strong foundation in statistical modelling. I am also a proficient coder: I released the python package “apc”, use R daily for my simulations, and taught MATLAB to grad students. I recently completed Udacity's Machine Learning Engineer Nanodegree and I found my capstone project in deep learning an eye-opening experience that has strengthened my interests in pursuing a career in deep learning.
Prior to my studies at Oxford, I was a full-time paramedic. During this time, I experienced the invaluable benefits of teamwork with goal-orientation and accuracy in situations of high pressure.
As a DPhil candidate, I work on econometric theory, specifically on age-period-cohort models. Such models are used in many fields such as actuarial science, demography, economics, epidemiology or sociology. Examples for applications are insurance claim reserving, research on demographic change and mortality forecasting.
In addition to my DPhil studies, I am very interested in deep learning. As a capstone project for my recently completed Machine Learning Nanodegree, I built a two-stage model for image classification. The first stage is a binary Faster R-CNN, the second stage a standard CNN. In an application to turtle and tortoise species classification, the model outperformed a benchmark CNN: top-1 and top-3 error improved by 1.25pp and 2pp, respectively. It was reviewed as “[…] one of the most detailed and well formed reports, on actual novel research, that I’ve seen as a capstone”.
'Over-dispersed age-period-cohort models', J. Harnau & B. Nielsen, Journal of the American Statistical Association.