If you know simulation based calibration checking (SBC), you will enjoy our new paper "Posterior SBC: Simulation-Based Calibration Checking Conditional on Data" with Teemu Säilynoja, @marvinschmitt.com and @paulbuerkner.com
https://arxiv.org/abs/2502.03279 1/5
=> More informations about this toot | More toots from avehtari@bayes.club
The original SBC checks whether the inference works for all possible data sets generated using the model and parameter draws from the prior. Priors are usually wider than posteriors and may contain regions where the computation fails. Illustration: Regions 1 and 3 exhibit bias in opposite directions, while inference is well calibrated within region 2. Prior SBC will not suggest calibration issues, while posterior SBC can assess inference for a posterior contained in one of the regions. 2/5
=> More informations about this toot | More toots from avehtari@bayes.club
For example, for hierarchical models, MCMC can have problems either with centered or non-centered parameterization depending on the data. Given one of the parameterizations, prior SBC observes both failing and non-failing inference. Posterior SBC focuses on the posterior conditional on the data, and can assess which parameterization works better for that specific data. 3/5
=> More informations about this toot | More toots from avehtari@bayes.club
We illustrate with a hierarchical normal and a Lotka-Volterra models using MCMC, and a drift diffusion model using amortized Bayesian inference. Posterior SBC is specifically useful for amortized inference, as the repeated inference has negligible cost. 4/5
=> More informations about this toot | More toots from avehtari@bayes.club
@MarvinSchmitt started collaborating on this while visiting Aalto University as @ELLISforEurope PhD student. ELLIS PhD student program has been great for increasing research visits and collaboration! 5/5
=> More informations about this toot | More toots from avehtari@bayes.club
text/geminiThis content has been proxied by September (3851b).