Approximate the Simulator, Not the Inference: Using Converging Hamiltonian Monte Carlo for Flexible SBI in Astronomy

Add to your list(s) Download to your calendar using vCal

• Matthew O'Callaghan (IoA, Cambridge)
• Monday 10 February 2025, 16:00-17:00
• Martin Ryle Seminar Room, KICC.

If you have a question about this talk, please contact km723.

Reliably including computationally intensive simulators with intractable densities in an inference problem is of paramount importance in astronomy. In this talk, I present a normalizing flow architecture that can replace intractable simulators in Bayesian models, while guaranteeing fast convergence of the No U-Turn Sampler variant of Hamiltonian Monte Carlo (HMC) to the posterior distribution. I discuss the limitations of using flow-based methods in HMC , and illustrate examples of a correct implementation for toy examples. I conclude by illustrating its use for estimating interstellar extinction from stellar colours.

This talk is part of the Astro Data Science Discussion Group series.

This talk is included in these lists:

• Astro Data Science Discussion Group
• Cambridge Astronomy Talks
• Combined External Astrophysics Talks DAMTP
• Cosmology, Astrophysics and General Relativity
• Institute of Astronomy Extra Talks
• Institute of Astronomy Talk Lists
• Martin Ryle Seminar Room, KICC

Note that ex-directory lists are not shown.