Skip to contentSkip to filters

We use cookies for analytics.

Cost-Aware Simulation-Based Inference

Ayush Bharti

Aalto University

Recording

50 min

Abstract

Simulation-based inference (SBI) is the preferred framework for estimating parameters of intractable models in science and engineering. A major challenge is the large computational cost of simulating data from complex models, a cost that often depends on the parameter values themselves.

This talk presents cost-aware SBI methods that substantially reduce the cost of existing sampling-based SBI methods such as neural SBI and approximate Bayesian computation. They combine rejection sampling with self-normalised importance sampling to cut the number of expensive simulations needed. The approach is studied extensively on models from epidemiology to telecommunications engineering, with large reductions in the overall cost of inference (Bharti, Huang, Kaski and Briol, arXiv:2410.07930).

Topics