Projects

Projects, who doesn’t like them? Unfortunately some of them are not easy to tackle alone. This list is a loose collections of projects I would love to collaborate on or things that I have done. Maybe I will use Jekyll Posts or make separate pages in the future. Shoot me an email if you find any of them interesting.

Precipitation Nowcasting with Deep Learning

For my MSc Thesis I developed a UNet with Pytorch and Pytorch-Lightning that forecasts the precipitation distribution over northern Germany 1h in advance. It uses the German Weather Service’s 1h-cumulative precipitation data from the RADOLAN Routine at t-5h, t-4h, t-3h, t-2h, t-1h, t. The output is for t+1h.

You can read the thesis here or here and find the code and some more details here

Although it was a proof of concept the model is still able to abstract the spatial features of the precipitation map. Here are two examples

missing — Figure: model input (6 top squares), model output (bottom left) and ground truth (bottom right) for 2020-02-09 19:50. The four different colors represent the four different precipitation classes' lower limits.

ECG VFib Alarm

Ventricular fibrillation (VFib) is a dangerous arrhythmia. Patients at risk have a (Subcutaneous-)Implantable Cardioverter Defibrillator (S-ICD) or use a Wearable Cardioverter Defibrillator (WCD). After a Myocardial Infarction (MI) the risk of VFib is increased, however this doesn’t necessarily lead to the implantation of an ICD.

Idea: use an ECG patch for constant patient telemetry and have an alarm ring for family members if a dangerous arrhythmia is detected. Obviously there needs to be an Automatic External Defibrillator (AED) nearby for this to have sense.

A Collection of Risk Calculators

Every time you want to compute your risk of a certain, illness, cancer or natural disaster you have to Google for the appropriate online tool. A collection on a good-looking website would be a stimulating side-project.