Event Description
Content: This tutorial introduces Simulation-Based Inference (SBI), a framework combining Bayesian modeling, AI techniques, and high-performance computing (HPC) to address key challenges, such as performing reliable inference with limited data by using AI-based approximate Bayesian computation. Moreover, it tackles the problem of intractable likelihood functions, thereby allowing to utilize Bayesian inference for biological systems with multiple sources of stochasticity. The tutorial also demonstrates how to leverage HPC environments to drastically reduce inference runtimes, making it highly relevant for large-scale biological problems. This tutorial bridges theoretical foundations with hands-on applications realized via jupyter notebooks.
Learning Objectives
Understand the Principles of Simulation-Based Inference (SBI): learn the theoretical foundations of SBI, including its relationship with Bayesian inference and its advantages in handling complex systems. Explore SBI Methods (SNPE, SNLE, and SNRE): gain an understanding of Sequential Neural Posterior Estimation (SNPE), Sequential Neural Likelihood Estimation (SNLE), and Sequential Neural Ratio Estimation (SNRE) and their applications.
Learn how to design and implement SBI frameworks for representative scenarios, such as molecular dynamics, cell growth, count data modeling, and Lotka-Volterra systems.
Leverage HPC for SBI Workflows: understand how to use high-performance computing (HPC) environments to scale SBI workflows and efficiently distribute computational workloads.
Prerequisites: Although the course is giving a brief introduction into Bayesian statistics and AI methods involved in building an SBI framework, we also expect basic familiarity with statistical and deep learning concepts. Experience of working with HPC systems would be beneficial but is not strictly required.
Target audience: Scientists who are willing to speed up their Bayesian inference methods using AI-based tools and simulations. Scientists who are willing to take their Bayesian inference to the next level by handling intractable likelihoods. Scientists who are willing to enhance their simulations with AI-based inference methods for uncertainty quantification.
Learning outcome: The ability to set up a Bayesian approach within a given framework
Language: This course is given in English.
Duration: 2 half days
Date: 7-9 September 2026, 13:00 - 17:00
Venue: Online
Number of Participants: maximum 25
Instructor: Alina Bazarova, Jose Robledo