Simulation-Based Inference
A class of statistical inference methods that require no explicit likelihood function. By sampling parameter-data pairs from a simulator, neural networks are trained to approximate posterior distributions or likelihood ratios.
Timeline (last 90 days)
In simulation-based inference (SBI), high-fidelity simulations are extremely expensive, and under a fixed budget, using an inaccurate simulator leads to a biased posterior distribution.
The paper demonstrates the method's practical performance in two scenarios: the distribution shift problem in simulation-based inference (SBI), and nonlinear independent component analysis (Nonlinear ICA).
All Facts (2)
In simulation-based inference (SBI), high-fidelity simulations are extremely expensive, and under a fixed budget, using an inaccurate simulator leads to a biased posterior distribution.
50%UnverifiedThe paper demonstrates the method's practical performance in two scenarios: the distribution shift problem in simulation-based inference (SBI), and nonlinear independent component analysis (Nonlinear ICA).
50%