Introduction to Simulation-based inference & SBI Clinic

Simulators are indispensable for modeling complex systems, from physical phenomena to industrial processes. But how do you determine the right parameters to make your simulations match observed data or predict future outcomes? This course introduces Simulation-based Inference (SBI), a recent approach that makes parameter estimation for simulators accessible and efficient. SBI enables applying the powerful framework of Bayesian inference to simulators, even when traditional methods fall short. By leveraging advances in probabilistic machine learning and generative modeling, SBI allows you to infer parameters directly from simulations without needing explicit likelihood functions. This two-day course is designed for professionals in research and industry who use simulators or need to perform parameter inference. Day 1 will provide a practical introduction to SBI. On day 2, participants will have the chance to apply SBI to their models, supervised by experts in the field. Through interactive examples and hands-on exercises, you’ll learn how to use SBI to tune your models, interpret results, and make data-driven decisions with confidence. Who is the event for? Researchers and engineers from all disciplines that use any sort of simulation-based model What prior knowledge is required for the event? Basic probability theory, basic Bayesian statistics, proficiency in Python, some experience in using ML methods and PyTorch. What are the intended learning objectives of the event? Understand the Principles of Simulation-based Inference (SBI): Acquire a thorough understanding of the theoretical foundations and practical motivations for using SBI as an alternative to traditional likelihood-based inference techniques. Learn Neural Approaches to Density and Posterior Estimation: Gain insight into neural network-based methods for density and posterior estimation, focusing on their implementation, advantages, and potential limitations. Evaluate Approximation Quality: Learn to assess the quality of inferred approximations using established metrics, such as posterior predictive checks, simulation-based calibration and the Classifier-2-Sample-Tests. Apply SBI to Practical Problems: Develop the skills to model, simulate, and perform inference on data in real-world scenarios using SBI techniques, emphasizing practical applications and implementation. Timetable: Tue: 09:00am – 05:00pm Wed: 09:00am – 03:00pm