Verified Expert in Engineering
Data Scientist and Developer
Daniel has 10+ years of experience in scientific analysis, machine learning, and programming, with a particularly strong foundation in (Bayesian) statistics. He places emphasis on rapid prototyping but can also deploy models all the way to production. His diverse background enables him to attack new problems from multiple perspectives—he has worked as an analyst, consultant, software engineer, as well as a researcher in neuroscience, biophysics, and discrete mathematics.
Google Cloud, Python, Pandas, NumPy, SciPy
The most amazing...
...algorithm I've invented improved depression relapse detection by over 5% compared to the state of the art that was established just six months before.
Expert Data Scientist
- Developed a synthetic patient case generator based on symptoms-disease networks. The project enabled new downstream analyses, such as demonstrating the effectiveness of an active learning model before deploying it to the customer.
- Created methods for forecasting individual disease risk. Implemented risk predictors for diabetes and cardiovascular diseases.
- Co-developed an NLP pipeline for extracting medical information from scientific articles (diseases, symptoms, risk factors, and epidemiological information) and implemented corresponding information architecture.
- Investigated the potential use of polygenic risk scores in improving individual disease risk prediction. Planned and executed experiments for evaluating polygenic risk scores via UK Biobank data.
- Organized bi-weekly meetings to exchange knowledge about AI and data science topics across the company.
- Mentored other data scientists across the company.
- Aggregated and curated multiple biomedical ontologies into one coherent knowledge graph.
- Developed a knowledge graph model and convolutional inference scheme for inference on a biomedical network graph constructed from several ontologies.
- Applied the inference method for predicting side effects of recently approved drugs.
Prediction of Relapse in Depression
This seemingly simple statement cannot be verified with the standard machine learning toolkit, so I developed a new method based on Gaussian processes to verify the hypothesis. Not only did this improve the current state of the start, which was just six months old at the time by more than 5%, but it also yielded various important medically directly interpretable insights into potential causes for relapse and how the influence of these causes might change over time.
Pandas, NumPy, SciPy, TensorFlow, PyTorch, SpaCy, Natural Language Toolkit (NLTK)
Bayesian Machine Learning, Probabilistic Graphical Models, Time Series Analysis, Neuroscience, Machine Learning, Deep Learning, Generalized Linear Model (GLM), Reinforcement Learning, Bayesian Inference & Modeling
PhD in Biomedical Engineering
ETH Zürich - Zürich, Switzerland