Scientific Focus
My lab develops interpretable machine learning methods to uncover the temporal dynamics underlying brain function and mental health. We work with multimodal neurophysiological and behavioural data (EEG, fMRI, spike trains, behaviour) to learn mechanistic models that can inform clinical decision-making in psychiatry, from predicting treatment response to developing new therapeutic approaches.
Methods
We use machine learning to learn dynamical systems models from time series data, with three interconnected priorities guiding our approach.
First, we emphasize mechanistic interpretability: we develop models that reveal the underlying principles that generate observed dynamics, such as network connectivity or computational strategies in brain recordings.
Second, we pursue multimodal integration of diverse data types.
Third, we focus on cross-subject generalization, developing methods that can integrate data and extract interpretable shared structure across individuals.
Selected Publications
Brenner, M., Weber, E., Koppe, G., Durstewitz, D.: Learning Interpretable Hierarchical Dynamics Models from Time Series Data, ICLR 2025
Brenner, M., Hemmer, C., Monfared, S., Durstewitz, D.: Almost-Linear RNNs Yield Highly Interpretable Symbolic Codes in Dynamical Systems Reconstruction, NeurIPS 2024
Brenner, M., Hess, F., Koppe, G., Durstewitz, D.: Integrating Multimodal Data for Joint Generative Modeling of Complex Dynamics, ICML 2024
Brenner, M., Hess, F. et al., Tractable Dendritic RNNs for Reconstructing Nonlinear Dynamical Systems, ICML 2022
Göring, N., Hess, F., Monfared, Z., Brenner, M., Durstewitz, D.: Out-of-Domain Generalization in Dynamical Systems Reconstruction, ICML 2024