Simulation-Based Inference

We provide a practical guide for applying SBI methods, outlining a structured workflow with guidelines and diagnostic tools for every stage—from setting up simulators to validating results.

This talk introduces a novel approach that bridges SBI and Pyro to enable simulation-based hierarchical Bayesian inference, combining SBI’s ability to handle intractable simulators with Pyro’s expressive power for complex hierarchical structures.

A comprehensive hands-on tutorial at EuroSciPy 2025 teaching scientists and engineers how to use simulation-based inference (SBI) for Bayesian parameter estimation in complex simulators, providing full uncertainty quantification beyond simple best-fit approaches.

We present sbi reloaded, a comprehensive update to the sbi Python package that provides researchers with state-of-the-art algorithms and tools for simulation-based inference workflows across scientific domains.

We dissect the origins of wiring specificity in dense cortical connectomes, revealing how geometric and biological constraints shape synaptic connectivity patterns.

This talk covers SBI, a method for Bayesian parameter inference using simulated data, and introduces the sbi library, an open-source tool for practitioners and researchers.

Internal seminar at TransferLab discussing recent advances in simulation-based inference methods and their practical applications across scientific domains.

We demonstrate how simulation-based inference can efficiently identify synaptic connectivity rules in dense cortical connectomes, enabling analysis of complex brain circuit organization.

Invited presentation at the Bernstein Center for Computational Neuroscience Berlin on using SBI methods for parameter inference in models of neural circuits and behavior.

We introduce GATSBI, a method that combines generative adversarial networks with simulation-based inference to achieve better sample efficiency and accuracy in posterior estimation.

We present a benchmark suite for simulation-based inference, systematically evaluating different methods across tasks with varying dimensionality, simulation budgets, and amortization requirements.

We introduce sbi, a Python package providing a unified interface for simulation-based inference methods, making these powerful techniques accessible to researchers across disciplines.

sbi is a Python package for simulation-based inference, providing a user-friendly interface to perform Bayesian parameter inference for simulator-based models with intractable likelihoods.