Ulrich Schaechtle

PhD Computer Science




Judea Pearl's seminal work on a mathematical framework for causal inference had a wide impact on the AI community. A computational understanding of causality has gained more and more interest since: the community understood that producing rich causal models computationally can be key to creating human-like artificial intelligence. Causal discovery seeks to develop algorithms that learn the structure of causal relations from observation. Diverse problems in domains such as aeronautical engineering, social sciences and bio-medical databases have acted as motivating applications that have gained new insights by applying suitable algorithms to large amounts of non-experimental data, typically collected via the internet or saved in public and private databases.



Probabilistic Programming

For sophisticated causal models, crafting sound and efficient inference algorithms becomes very difficult as the necessary derivations often involve cumbersome algebra and non-trivial proofs of convergence. As a result, new models are usually communicated using a mix of natural language, pseudo-code and mathematical formulae. Inference is specific to its purpose and often not re-usable. Designing, implementing and testing inference algorithms is hard even for experts. Small changes in requirements or assumptions can result in a change of the model or affect the inference algorithm. This can cause a lot of work for an expert and can keep non-experts outside of the domain of machine learning and causal inference, completely, since they may lack the necessary mathematical background. Probabilistic programming seeks to overcome these shortcomings with two main features:

  1. an expressive language able to accommodate both declarative and procedural semantics to represent probabilistic knowledge and dependencies in a sound way; and
  2. a generalised inference engine that mitigates the drawbacks of one-off inference algorithm design.




I am interested in implications and applications of my research in the domains of Biology and Medicine. I was part of COMMODITY12, an international board of experts developing a personal health system to assist in the provision of continuous and personalised health services to diabetic patients. The system consists of ambient and/or body (wearable and portable) devices, which acquire, monitor and communicate physiological parameters and other health-related context of an individual, such as physical activity and vital body signals.
My role in this project was to provide machine learning algorithms, taking into account expert biomedical knowledge and rich data structures to derive important insights about the individual’s health status given large sets of non-experimental data.