Act-i-ML

Act-i-ML, active informed machine learning, is a joint project between Free University Berlin and Hasso Plattner Institute for Digital Engineering, Potsdam that is funded by the BMBF. Therein, the groups of Bernhard Renard and Katharina Baum work together to develop resilient methods for the use of AI. In particular, we aim to improve on a branch of machine learning that includes prior knowledge, e.g. in the form of differential equations, to counteract issues with small training data. This setting is frequent e.g. during epidemic forecasting where only few timepoints are available, or in semiconductor ion flux computations where measurements and explicit numerical computations are expensive. We are building on our previous work, SimbaML [1], and our investigations on synthetic data properties required for successful simulation-based transfer learning [2].

We are excited for our first project members to start their work on January 1, 2025! Our kick-off meeting will take place at the Hasso Plattner Institute on January 16, 2025.

[1] Kleissl M, Drews L, Heyder BB, Zabbarov J, Iversen P, Witzke S, Renard BY, Baum K. SimbaML: Connecting mechanistic models and machine learning with augmented data. ICLR 2023 Tiny Paper https://openreview.net/forum?id=1wtUadpmVzu
[2] Zabbarov J, Witzke S, Kleissl M, Iversen P, Renard BY, Baum K. Optimizing ODE-derived Synthetic Data for Transfer Learning in Dynamical Biological Systems. biorxiv preprint, https://doi.org/10.1101/2024.03.25.586390

This project is funded by the BMBF, and it is supported by the DLR.