Mind the Gap: Methods and Applicability of Simulation-Based Inference

Jan Teusen (né Boelts), AI Researcher at appliedAI Institute, will give a gentle introduction to simulation-based inference (SBI), along with a presentation of his own research on new SBI methods for computational neuroscience and new software tools.

Abstract

In science and engineering, we often use computer simulations as models of real-world phenomena. Such simulation-based models allow us to build and test hypotheses about the processes underlying a phenomenon, e.g., by simulating synthetic data from a model and comparing it to observed data. A key challenge with this approach is finding model configurations that reproduce the observed data.

Bayesian statistical inference provides a principled way to address this challenge: It allows us to systematically infer suitable model parameters and quantify uncertainty. However, classical Bayesian inference methods typically require access to the likelihood function of the model and thus cannot be applied to the often highly complex scientific simulators.

With the increase in available computational resources and the advent of neural network-based machine learning methods, an alternative approach has recently emerged: simulation-based inference (SBI). SBI enables Bayesian parameter inference but only requires access to simulations from the model.

In this talk, I give a gentle introduction to SBI, present my research on new SBI methods for computational neuroscience, and comment on new software tools bridging the gap between methods and applicability of SBI.

References

[Cra20F]

The frontier of simulation-based inference, Kyle Cranmer, Johann Brehmer, Gilles Louppe.

Dec 2020

Many domains of science have developed complex simulations to describe phenomena of interest. While these simulations provide high-fidelity models, they are poorly suited for inference and lead to challenging inverse problems. We review the rapidly developing field of simulation-based inference and identify the forces giving additional momentum to the field. Finally, we describe how the frontier …

[Pap18F]

Fast $\epsilon$-free Inference of Simulation Models with Bayesian Conditional Density Estimation, George Papamakarios, Iain Murray.

Apr 2018

Many statistical models can be simulated forwards but have intractable likelihoods. Approximate Bayesian Computation (ABC) methods are used to infer properties of these models from data. Traditionally these methods approximate the posterior over parameters by conditioning on data being inside an $\epsilon$-ball around the observed data, which is only correct in the limit …

[Tej20S]

sbi: A toolkit for simulation-based inference, Alvaro Tejero-Cantero, Jan Boelts, Michael Deistler, Jan-Matthis Lueckmann, Conor Durkan, Pedro J. Gonçalves, David S. Greenberg, Jakob H. Macke.

Aug 2020

Scientists and engineers employ stochastic numerical simulators to model empirically observed phenomena. In contrast to purely statistical models, simulators express scientific principles that provide powerful inductive biases, improve generalization to new data or scenarios and allow for fewer, more interpretable and domain-relevant parameters. Despite these advantages, tuning a simulator’s …

Abstract

References

In this series →