Ongoing Projects
Regressing the age of bones
Estimating the so-called bone-age in a hand's radiograph is relevant for different diagnostics in medicine. Current approaches rely either on a lengthy manual process or on uninterpretable deep learning approaches.
In this project, we aim to describe simple interpretable features that allow a doctor to estimate bone age following a simple linear regression. We use deep learning to automatically extract such features.
Reinforcement learning and word equations
The problem of solving word equations (also known as string equations) is an NP-hard (conjectured NP-complete) problem used in software verification, database querying, and more. It is also of high theoretical relevance within the areas of computer science, group theory, and logic. See here (applied) and here (theoretical) for surveys on this topic.
In this ongoing project we are building a non-complete string solver (i.e. word equation solver) which leverages deep reinforcement learning and other existing string solvers. See the corresponding repository and preprint.
Secret project
It's a secret (for now) :D