Abstract

Simulation-based inference (SBI) has emerged as a powerful framework for performing Bayesian inference when likelihood functions are intractable. However, there remains a significant gap between methodological advances and practical deployment. In this talk, I discuss:

  1. Current State of SBI Methods: Overview of neural posterior/likelihood/ratio estimation approaches
  2. The Applicability Gap: Why aren’t more practitioners using SBI?
  3. Software Solutions: How the sbi package addresses usability challenges
  4. Real-world Case Studies: Applications in neuroscience, epidemiology, and climate science
  5. Future Directions: Scaling SBI to higher dimensions and real-time applications

Key Takeaways

  • SBI methods are mature enough for practical deployment
  • Software accessibility is crucial for adoption
  • Domain-specific customization often required
  • Diagnostics and validation remain challenging

Venue

TransferLab Internal Seminar Series
appliedAI Institute for Europe
Munich, Germany