With the surge of probabilistic modeling, more and more branches of science subscribe to the use of Bayesian models for inference and uncertainty quantification. In recent years, deep learning has proven indispensable for scaling up Bayesian inference to challenging inverse problems in the life sciences. One such problem is determining the hidden parameters of complex, biophysically detailed models from high-dimensional observations, such as neuroimaging or microscopy data. With the help of Bayesian simulatin-based inference (SBI), we are building the most detailed full-head fNIRS simulator and developing hierarchical Bayesian approaches for fluorescence lifetime imaging (FLIM). Moreover, we are pushing the BayesFlow project for AI-assisted SBI, which enables us to solve a variety of computationally challenging tasks using generative AI.

Lead PI: Dr. Stefan T. Radev