Jan Teusen (né Bölts)

A central challenge in many areas of science and engineering is to identify model parameters that are consistent with prior knowledge and empirical data. Bayesian inference offers a principled framework for this task, but can be computationally prohibitive when models are defined by stochastic simulators. Simulation-based Inference (SBI) is a suite of methods developed to overcome this limitation, which has enabled scientific discoveries in fields such as particle physics, astrophysics, and neuroscience. The core idea of SBI is to train neural networks on data generated by a simulator, without requiring access to likelihood evaluations. Once trained, inference is amortized: the neural network can rapidly perform Bayesian inference on empirical observations without requiring additional training or simulations. In this tutorial, we provide a practical guide for practitioners aiming to apply SBI methods. We outline a structured SBI workflow and offer practical guidelines and diagnostic tools for every stage of the process—from setting up the simulator and prior, choosing and training inference networks, to performing inference and validating the results. We illustrate these steps through examples from astrophysics, psychophysics, and neuroscience.

The sbi package provides a unified Python interface for simulation-based inference (SBI) algorithms, enabling researchers to perform Bayesian inference on models with intractable likelihoods across diverse scientific domains. This paper presents sbi reloaded, a comprehensive update that has evolved through extensive community contributions and user feedback. The package now includes multiple state-of-the-art inference algorithms (NPE, NLE, NRE, FMPE, MNLE, and more), enhanced diagnostics and validation tools, improved computational efficiency, and a flexible API supporting custom neural network architectures and proposal distributions. Key improvements include GPU acceleration, distributed training support, comprehensive testing infrastructure, and extensive documentation with tutorials. With over 30 contributors and a growing user base, sbi has become the standard toolkit for neural simulation-based inference, balancing ease of use with the flexibility needed for cutting-edge research.

Understanding the organizational principles of neural circuits requires identifying the rules governing synaptic connectivity between neurons. However, inferring these rules from experimental data is challenging due to the complexity of biological networks and limitations in data acquisition. Here, we present a simulation-based inference approach for identifying generative models of connectivity in anatomically detailed cortical circuits. Our method combines biologically realistic simulations of cortical connectivity with neural density estimation to perform Bayesian inference on connectivity parameters. We validate our approach on synthetic data and apply it to a dense reconstruction of rat barrel cortex, revealing distance-dependent connectivity rules and cell-type-specific biases. The method is computationally efficient, requiring orders of magnitude fewer simulations than traditional approaches.

Inferring parameters of computational models that capture experimental data is a central task in cognitive neuroscience. Bayesian statistical inference methods usually require the ability to evaluate the likelihood of the model—however, for many models of interest in cognitive neuroscience, the associated likelihoods cannot be computed efficiently. Simulation-based inference (SBI) offers a solution to this problem by only requiring access to simulations produced by the model. Here, we provide a new SBI method that is substantially more simulation-efficient. Our approach, Mixed Neural Likelihood Estimation (MNLE), trains neural density estimators on model simulations to emulate the simulator, and is designed to capture both the continuous (e.g., reaction times) and discrete (choices) data of decision-making models. We demonstrate MNLE on two variants of the drift-diffusion model (DDM) and show that it is substantially more efficient than LANs, achieving similar likelihood accuracy with six orders of magnitude fewer training simulations.