BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-6856
DTSTART:20220216T170000Z
DTEND:20220216T181500Z
DTSTAMP:20260414T154050Z
LAST-MODIFIED:20220208T212508Z
LOCATION:Chamberlin 5280 (Zoom link also available for online particip
 ants who signed up on our mailing list)
SUMMARY:Fast and Credible Inference with Truncated Marginal Neural Rat
 io Estimation\, Physics ∩ ML Seminar\, Alex Cole \, University of Am
 sterdam
DESCRIPTION:Across fields\, scientific models are computationally impl
 emented via parametric stochastic simulators. However\, solving the �
 �inverse problem” and constraining model parameters from data is a c
 hallenge in this context. Recently\, the field of simulation-based inf
 erence has made great strides thanks to deep learning methods. I will 
 outline a new method in simulation-based inference called Truncated Ma
 rginal Neural Ratio Estimation (TMNRE). TMNRE is (i) simulation-effici
 ent\, actively identifying the relevant regime of parameter space with
 out sacrificing amortization (ii) scalable to high-dimensional data an
 d model parameter spaces (iii) trustworthy\, in the sense that statist
 ical consistency tests beyond those available to e.g. MCMC can be rapi
 dly performed. I will show examples of these benefits in the context o
 f cosmological inference. I will also describe our development of a us
 er-friendly and general package for TMNRE called swyft.
\n
\nImplement
 ation of TMNRE available at https://github.com/undark-lab/swyft. Talk 
 based on https://arxiv.org/abs/2011.13951 (NeurIPS ML4PS ’20)\, http
 s://arxiv.org/abs/2107.01214 (NeurIPS ’21)\, https://arxiv.org/abs/2
 111.08030 .
URL:https://www.physics.wisc.edu/events/?id=6856
END:VEVENT
END:VCALENDAR