Multi-fidelity ABC #496
Description
Feature description
Implement the ability to use low-fidelity models with separate acceptance criterion, to determine whether to simulate the costly high-fidelity model via early rejection/acceptance, and thus overall reduce the required number of high-fidelity model simulations.
E.g. based on:
- Multifidelity Approximate Bayesian Computation
- Multifidelity Approximate Bayesian Computation with Sequential Monte Carlo
Parameter Sampling
Requires:
- Definition of low-fidelity simulators
- Separate acceptance threshold schemes and distance functions. Can adaptive schemes be used, if e.g. not all particles have high-fidelity simulations?
- Storage extension
- Potentially low-fidelity templates for common models, such as discretization, tau leaping, ML approaches (require separate training)
Motivation/Application
Reduce overall required number of costly high-fidelity simulations by determining early whether a parameter is promising.