I am a PhD candidate at the Max Planck International Research School for Intelligent Systems and the University of Tübingen. I am part of the Machine Learning in Science lab led by Jakob Macke. I work on Machine Learning tools for scientific discovery, in particular in neuroscience. Below, I outline my main research interests.
Biophysical modelling allows detailed insights into the processes underlying neural activity. I develop tools and methods that enable to build large-scale biophysical models and to tune them such that they match measurements and perform computations.
Differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics
Deistler, Kadhim, Beck, Pals, Huang, Gloeckler, Lappalainen, Schröder, Berens, Gonçalves, Macke (bioRxiv)
Diffusion Tempering Improves Parameter Estimation with Probabilistic Integrators for Ordinary Differential Equations
Beck, Bosch, Deistler, Kadhim, Macke, Hennig, Berens (ICML)
Neural circuits are remarkably energy-efficient and robust to perturbations. I use computational modelling and experimental measurements to study the biophysical properties underlying these features.
Energy efficient network activity from disparate circuit parameters
Deistler, Macke‡, Gonçalves‡ (PNAS)
Training deep neural density estimators to identify mechanistic models of neural dynamics
Gonçalves†, Lueckmann†, Deistler†, Nonnenmacher, Öcal, Bassetto, Chintaluri, Podlaski, Vogels, Greenberg, Macke (Elife)
Deep inverse modeling reveals dynamic-dependent invariances in neural circuit mechanisms
Gao, Deistler, Schulz, Gonçalves, Macke (bioRxiv)
Combined statistical-mechanistic modeling links ion channel genes to physiology of cortical neuron types
Bernaerts, Deistler, Goncalves, Beck, Stimberg, Scala, Tolias, Macke, Kobak, Berens (bioRxiv)
Mechanistic models provide interpretable and causal explanations for the processes underlying measurements. I develop probabilistic machine learning methods which allow to tune mechanistic models such that they match measurements.
All-in-one simulation-based inference
Gloeckler, Deistler, Weilbach, Wood, Macke (ICML, oral)
Amortized Bayesian Decision Making for simulation-based models
Gorecki, Macke, Deistler (TMLR)
Generalized Bayesian Inference for Scientific Simulators via Amortized Cost Estimation
Gao†, Deistler†, Macke (NeurIPS)
Adversarial robustness of amortized Bayesian inference
Gloeckler, Deistler, Macke (ICML)
Truncated proposals for scalable and hassle-free simulation-based inference
Deistler, Gonçalves‡, Macke‡ (NeurIPS)
Efficient identification of informative features in simulation-based inference
Beck, Deistler, Bernaerts, Macke, Berens (NeurIPS)
Variational methods for simulation-based inference
Gloeckler, Deistler, Macke (ICLR, spotlight)
Group-equivariant neural posterior estimation
Dax, Green, Gair, Deistler, Schölkopf, Macke (ICLR)
Together with many other amazing people, I am developer and maintainer of the following two toolboxes:
sbi—a toolbox for simulation-based inference
Tejero-Cantero†, Boelts†, Deistler†, Lueckmann†, Durkan†, Gonçalves, Greenberg, Macke (Journal of Open Source Software)
† indicates shared first authorship.
‡ indicates shared last authorship.