I am Greg, a data scientist intern at Roche working on using deep learning to syntheise
clinical tabular data.
Prior to my placement at Roche, I completed my PhD at ETH Zurich investigating dynamics of alpine glaciers using geophysical methods. I have an extensive background in geophysics after working for 7 years in the hydrocarbon industry as a technical team leader in seismic data processing. Alongside my 7+ years of professional experience, I have a strong background in mathematics, statistics and programming, as well as project and people management.
For my PhD I focused on understanding the dynamics of glacier hydraulics using geophysical techniques.
For my bachelor thesis I performed logistic regression in order to calculate the probability of each outcome in UK football matches.
Leveraging machine learning to learn how fast I ride my road bike at different grades in order to predict estimated moving time of a new route.
Benchmarking variational autoencoders against generative adversial networks for syntheising clinical data.
This is my list of first author publications that I completed during my doctorate studies. For an exhaustive list of my publications head over to my Google Scholar profile
Englacial hydrology plays an important role in routing surface water to the glacier's bed and it consequently affects the glacier's dynamics. However, it is often difficult to observe englacial conduit conditions on temperate glaciers because of their short-lived nature. This journal publication demonstrates that with the use of a wide range of geophysical measurements, we are able to detect and characterise an englacial hydrological system. We use radar, sesimics and borehole measurements into the hydrological feature to define its geometry. From the observations, we infer that the englacial conduit network is fed by surface meltwater and morainal streams. The surface and morainal streams merge together, enter the glacier subglacially and flow through subglacial channels along the flank. These subglacial channels flow into highly efficient englacial conduits traversing the up-glacier section of the overdeepening before connecting with the subglacial drainage system.
Englacial conduits act as water pathways to feed surface meltwater into the subglacial drainage system. A change of meltwater into the subglacial drainage system can alter the glacier’s dynamics. Between 2012 and 2019, I performed repeated ground-penetrating radar (GPR) surveys over an active englacial conduit network. The GPR data were processed using an impedance inversion workflow to determine the conduit’s infill material. The spatial and temporal evolution of the impedance inversion provided insights into the morphology of the Rhonegletscher’s englacial conduit network. During the summer melt seasons, we observed an active, water-filled, sediment-transporting englacial conduit network. We speculate that extensional hydraulic fracturing is responsible for the formation of the conduit as a result of the conduit network geometry observed and borehole observations. During the winter periods, the englacial conduit no longer transports water and either physically closed or be came very thin. Furthermore, the englacial conduit reactivated during the following melt season at an identical position as in the previous year.
Hydrological systems of glaciers have a direct impact on the glacier dynamics. Since the 1950s, geophysical studies have provided insights into these hydrological systems. Unfortunately, such studies were predominantly conducted using 2D acquisitions along a few profiles, thus failing to provide spatially unaliased 3D images of englacial and subglacial water pathways. The latter has likely resulted in flawed constraints for the hydrological modelling of glacier drainage networks. Here, we present 3D ground-penetrating radar (GPR) results that provide high-resolution 3D images of an alpine glacier's drainage network. Our results confirm a long-standing englacial hydrology theory stating that englacial conduits flow around glacial overdeepenings rather than directly over the overdeepening. Furthermore, these results also show exciting new opportunities for high-resolution 3D GPR studies of glaciers.