Stefan Radev


My research follows two streams. The first focuses on developing new Bayesian methods through the emerging generation of generative neural networks. The second focuses on building and applying computational models of complex processes (e.g., cognition, disease outbreaks) to gain insights from data (and sometimes Big Data). These two streams converge at the BayesFlow framework for Bayesian inference with modern deep learning (GitHub)(project page), of which I am the core developer and maintainer. 

In the new computer age, modern Bayesian inference allows us to estimate, validate, and draw substantive conclusions from high-fidelity probabilistic models. Typical problems in Bayesian analysis are:

  1. Estimate the posterior distribution of hidden parameters from noisy data (i.e., inverse inference);
  2. Compare competing models in terms of their complexity and predictive performance;
  3. Emulate the observable behavior of a system or predict future behavior under uncertainty.

However, despite their theoretical appeal and utility, Bayesian workflows are limited by severe computational bottlenecks: Analyzing even a single data set may already eat up days of computation, so that model validation and calibration become completely infeasible.

BayesFlow addresses these challenges by training custom generative neural networks on model simulations. Researchers can then re-use and share these networks for any subsequent application of the model. Since the trained networks can perform inference almost instantaneously (typically well below one second), the upfront neural network training amortizes quickly. For instance, amortized inference allows us to test a model’s ability to recover its parameters or assess its uncertainty calibration for different data set sizes in a matter of seconds, even though this may require the estimation of thousands of posterior distributions.

If you are a student who wants to do research at the frontier of probabilistic modeling, drop me a mail or simply pass by my office.

Primary Research Focus
Deep Learning and Probabilistic Modeling


I teach courses revolving around computational modeling, probabilistic inference, and statistics.

Office Hours
  • Thursday 4:00 - 6:00PM
Current Courses
  • COGS 6960 / PSYC 4960: Cognitive Modeling

    Description: Computational models allow us to instantiate assumptions about cognitive mechanisms and information processing as randomized computer programs. This course takes a simulation-based approach to cognitive modeling and introduces participants to formal workflows for model development, estimation, and criticism. It will cover a broad range of existing model families and encourage participants to build their own models. Along the way, participants will learn best practices for Python workflows and fully Bayesian analysis. 


Presentations & Appearances
  • More expressive amortized Bayesian inference via joint learning and self-consistency? (2023). Talk at the One World Approximate Bayesian Computation (ABC) Seminar.  Virtual event (Department of Statistics, Warwick, UK).
  • JANA: Jointly amortized neural approximation of complex Bayesian models. (2023). Spotlight presentation and poster at the 39th Conference on Uncertainty in Artificial Intelligence (UAI 2023). Pittsburgh, PA, USA.
  • Principled amortized Bayesian inference with deep learning. (2023). Conference workshop at MathPsych/ICCM/EMPG 2023. Amsterdam, Netherlands.
  • Deep learning for cognitive modeling. (2023). Symposium organized at MathPsych/ICCM/EMPG 2023. Amsterdam, Netherlands.
  • Compressing Bayesian inference with information maximization. (2023). Talk at MathPsych/ICCM/EMPG 2023. Amsterdam, Netherlands.
  • I simulate, therefore I understand? Simulacrum and explanation. (2023). Talk at the Conference on Interdisciplinary Research in Philosophy and Psychology (AG Philosophie und Psychologie). Cologne, Germany.
  • One framework to learn them all: Amortizing Bayes’ rule. (2022). Talk at the Meeting of the European Mathematical Psychology Group (EMPG 2022). Roverto (TN), Italy.
  • BayesFlow: New advances from the frontier of simulation-based inference. (2022). Talk at the SIAM Conference on Uncertainty Quantification (UQ 2022). Atlanta, Georgia, USA.
  • BayesFlow: Scalable amortized Bayesian inference with invertible networks. (2020). Poster at NeurIPS Europe Meetup on Bayesian Deep Learning. Virtual event.
  • Amortized Bayesian inference for models of cognition. (2020). Talk and conference paper at MathPsych/ICCM 2020. Virtual event.


Recent Preprints

  • Müller, J., Kühmichel, L., Rohbeck, M., Radev, S. T., & Кöthe, U. (2024). Towards context-aware domain generalization: Understanding the benefits and limits of marginal transfer learning. arXiv preprint arXiv:2312.10107. (arXiv)
  • Schmitt, M., Pratz, V., Köthe, U., Bürkner, P. C., & Radev, S. T. (2023). Consistency models for scalable and fast simulation-based inference. arXiv preprint arXiv:2311.10671. (arXiv)
  • Schmitt, M., Radev, S. T., & Bürkner, P. C. (2023). Fuse it or lose it: Deep fusion for multimodal simulation-based inference. arXiv preprint arXiv:2311.10671. (arXiv)
  • Elsemüller, L., Olischläger, H., Schmitt, M., Bürkner, P. C., Köthe, U., & Radev, S. T. (2023). Sensitivity-aware amortized Bayesian inference. arXiv preprint arXiv:2310.11122. (arXiv)
  • Bockting, F., Radev, S. T., & Bürkner, P. C. (2023). Simulation-based prior knowledge elicitation for parametric Bayesian models. arXiv preprint arXiv:2308.11672. (arXiv)
  • Elsemüller, L., Schnuerch, M., Bürkner, P. C., & Radev, S. T. (2023). A deep learning method for comparing Bayesian hierarchical models. arXiv preprint arXiv:2301.11873. (arXiv)

Conference Proceedings

  • Schmitt, M., Habermann, D., Bürkner, P. C., Köthe, U., & Radev, S. T. (2023). Leveraging self-consistency for data-efficient amortized Bayesian inference. UniReps Workshop, NeurIPS, New Orleans. (arXiv)
  • Radev, S. T., Schmitt, M., Pratz, V., Picchini, U., Köthe, U., & Bürkner, P. C. (2023). JANA: Jointly amortized neural approximation of complex Bayesian models. Proceedings of the 39th Conference on Uncertainty in Artificial Intelligence (UAI), 216, 1695-1706. (arXiv)(PMLR)
  • Müller, J., Radev, S. T., Schmier, R., Draxler, F., Rother, C., & Köthe, U. (2023). Finding competence regions in domain generalization. Transactions on Machine Learning Research (TMLR). (arXiv)(OpenReview)
  • Schmitt, M., Bürkner, P. C., Köthe, U., & Radev, S. T. (2023). Detecting model misspecification in amortized Bayesian inference with neural networks. Proceedings of the 45th German Conference on Pattern Recognition (GCPR), 1-9. (arXiv)(TBA)
  • Schmitt, M., Radev, S. T., & Bürkner, P. C. (2022). Meta-uncertainty in Bayesian model comparison. In International Conference on Artificial Intelligence and Statistics (AISTATS), 11-29. (arXiv)(PMLR)
  • Radev, S. T., Voss, A., Wieschen, E. M., & Bürkner, P. C. (2020). Amortized Bayesian inference for models of cognition. In International Conference on Cognitive Modelling (ICCM) Conference Proceedings. (arXiv)(ICCM)

Accepted Papers

  • Bürkner, P. C., Scholz, M., & Radev, S. T. (2023). Some models are useful, but how do we know which ones? Towards a unified Bayesian model taxonomy. Statistics Surveys, (17), 216-310.
  • Radev, S. T., Schmitt, M., Schumacher, L., Elsemüller, L., Pratz, V., Schälte, Y., Köthe, U., & Bürkner, P. C. (2023). BayesFlow: Amortized Bayesian workflows with neural networks. Journal of Open Source Software, 8(89), 5702.
  • Schumacher, L., Bürkner, P. C., Voss, A., Köthe, U., Radev, S. T. (2023). Neural superstatistics for Bayesian estimation of dynamic cognitive models. Scientific Reports, (13), 13778.
  • von Krause*, M., Radev*, S. T., & Voss, A. (2022). Mental speed is high until age 60 as revealed by analysis of over a million participants. Nature Human Behaviour, 6(5), 700-708.
  • Radev S. T., D’Alessandro M., Mertens U. K., Voss A., Köthe U., & Bürkner P. C. (2021). Amortized Bayesian model comparison with evidental deep learning. IEEE Transactions on Neural Networks and Learning Systems, 1 -15.
  • Radev, S. T., Graw, F., Chen, S., Mutters, N. T., Eichel, V. M., Bärnighausen, T., & Köthe, U. (2021). OutbreakFlow: Model-based Bayesian inference of disease outbreak dynamics with invertible neural networks and its application to the COVID-19 pandemics in Germany. PLoS Computational Biology, 17(10), e1009472.
  • Bieringer, S., Butter, A., Heimel, T., Höche, S., Köthe, U., Plehn, T., & Radev, S. T. (2021). Measuring QCD splittings with invertible networks. SciPost Physics, 10(6), 126.
  • Radev, S. T., Mertens, U. K., Voss, A., Ardizzone, L., & Köthe, U. (2020). BayesFlow: Learning complex stochastic models with invertible neural networks. IEEE Transactions on Neural Networks and Learning Systems, 33(4), 1452-1466.

View All Scopus Publications

Back to top