Hi, I am Tobias Schanz
Currently diving deep into machine learning for my PhD in the
model-driven machine learning group at the Helmholtz-Zentrum Hereon, near Hamburg, Germany, I find myself at the crossroads of neural networks and earth system sciences. My quest? To pioneer training methods for generative neural networks that refine the statistical accuracy of predictions—both individually and collectively within an ensemble. This matters immensely as it directly influences our confidence in these predictions and, by extension, the decisions we make based on them.
Beyond the core of my research, I’m passionate about high performance computing, unveiling stories hidden in data through visualization, and tackling the challenges of big data. My teammates might say I’m the go-to for coding prowess, seamless project coordination, and bringing new ideas to life with speed and precision.
I'm looking forward to completing my PhD in June 2024 and am eager to explore opportunities in machine learning, deep learning, and data science thereafter. I'm especially keen on further developing deep neural networks, leveraging my extensive expertise in this area. While I thrive on challenging tasks that push the boundaries of technology, I'm seeking roles that go beyond simple data migration or cleaning up spreadsheets.
If you are interested in working with me, please reach out to me via email or LinkedIn.
PhD Student at Helmholtz-Zentrum hereon
2020 — July 2024
At the moment I am a PhD student at the Helmholtz-Zentrum hereon. My PhD is about finding new ways to train generative neural networks with the goal of controlling statistical properties of not only the predictions of each ensemble member but also of the ensemble as a whole. This is important because the predictions of the ensemble members are often used to estimate the uncertainty of the prediction. If the ensemble members are not diverse enough, the uncertainty estimate will be too low. This can lead to wrong decisions and potentially to wrong actions.
I also work on a project about semi-supervised learning, which I explain in more detail in the projects section.
Free Data Science Contractor
2020 — 2020
From a contact, which I got to know during my time at AKRA, I was asked to help a B2B company from Hamburg in integrating their data into a new management software. I was able to help them with the data integration and also with the development of a new data pipeline to structure their data in an efficient way.
Data Science Consultant at AKRA GmbH
2019 — 2020
After finishing my Masters I started working as a data scientist at AKRA. I worked on multiple projects for different customers, but my tasks mainly included data quality assurance, data analysis and visualization, in terms of dashboards and reports. The few machine-learning tasks I worked on were mostly simple clustering (DBSCAN) and outlier detection (Isolation Forest). During this time I learnt a lot about databases (InfluxDB, PostgreSQL) and object relational mapping (SQLAlchemy). I also took on tasks that did not directly involve data science, like creating simple APIs and such.
The main takeaway from my time at AKRA was how to communicate my results in a structured way, but I realized that I wanted to focus more on machine learning and less on data engineering.
M.Sc. Meteorology at the University of Hamburg
2017 — 2019
During my master studies I was given a lot more freedom in the courses to take than during the bachelors and so I found myself sitting in a lot of high performance computing and big data lectures. This led me to write my master thesis about the application of convolutional neural networks for processing raw radar measurements into radar echo maps. I found a way to train the neural network on synthetic data and then apply it to real world data. In the end, my model even outperformed the operational algorithm from the University of Hamburg.
The left image shows the result of my CNN, the right image shows the result of the operational algorithm. The CNN is able to resolve the structure of the precipitation much better than the operational algorithm.
Research Cruise Over the Pacific Ocean
2019 — 2019
I had the chance to participate in a six week long research cruise over the pacific ocean from Vancouver (Canada) to Singapore, as a student helper for the Max-Planck-Institute for Meteorology. There were two projects on board of the ship "Sonne". The first was about microplastic in the ocean, the second was about onboard radiation measurements. I was part of the radiation team and we were able to collect a lot of useful data that could later be used to verify satellite measurements. An overview of the cruise can be found here.
Max-Planck-Institute for Meteorology
2016 — 2019
Nearly all the time during my studies, I was working as a student helper at the Max-Planck-Institute for Meteorology. I started out doing smaller tasks like optimizing Python routines in terms of speed and memory usage. Later on I was able to work on my own projects, was in charge for data integrity and developed a lot of useful tools for the scientists. Check out some of them in the projects section.
B.Sc. Meteorology at the University of Hamburg
2014 — 2018
I was a student at the University of Hamburg from 2014 to 2018. I studied meteorology and wrote my bachelor thesis about a new retrieval method for the estimation of water vapor in the atmosphere by using a combination of a radiation model (PSRad) and surface based measurements from the NASA (AERONET). To be fair, the retrieval did not work very well but I learnt how to set up a research project, how to work with unstructured data, how a radiation model works and how to write a scientific report.
Using different atmospheric profiles in the radiation model for the retrieval, I show how my algorithm performs compared to direct measurements from NASA. BSRN is my algorithm, AERONET is the NASA measurement.
Open to Work from August 2024 
PhD Student at Helmholtz-Zentrum hereon
