Projects
At BayesOps, we develop robust software and novel neural architectures for end-to-end Bayesian modeling. The lab is generously supported by the National Science Foundation (NSF).
Highlighted Projects
BayesFlow is a Python library for simulation-based inference with generative AI. It provides users and researchers with a user-friendly API for rapid Bayesian workflows, a rich collection of neural architectures, and multi-backend support via Keras3.
We are developing neural networks that will allow researchers to fit thousands of cognitive models. The project’s focus on robust inference and benchmarking aims to strenghten the emerging infrastructure for model-based analysis in the behavioral sciences.
CogFormer is the first incarnation of the BayesGPT project, allowing cognitive modelers to perform near-instant Bayesian modeling on a combinatorial space of design configurations. The web app lets you interact with a pre-trained version of CogFormer v1.
Score-based diffusion models enable fast and accurate simulation-based inference. Our tutorial synthesizes recent developments, covering design choices for training, sampling, and evaluation, as well as the inherent flexibility of scores for adaptive inference.
Bayesify is a multi-stage agentic AI framework for evaluating the integrity of Bayesian workflows. Backed by curated rubrics from the methodological literature, it grades your paper step by step and helps you bring your workflow to the gold standard.
Applications
In a first-of-its-kind application, we use simulation-based inference for modeling human motion while experiencing shared virtual worlds, paving the way for model-based research on human interaction with the built environment.