Simulation-based inference

tags
Markov Chain Monte Carlo

Algorithm based on that intuition from Rubin (?) that drawing samples from the posterior distribution was equivalent to simulate data from the prior predictive distribution and accept or reject the sample based on the distance to the observed data. You need two ingredients:

Algorithms

  • Approximate Bayesian Computation n
  • Sequential Neural Posterior Estimation from
  • Sequential Neural Likelihood Estimation
  • Sequential Neural Ratio Estimation
    • ALLR
    • SRE

      \begin{equation}
        \mathcal{F} = \int_{0}^{\infty} f(x) dx
      \end{equation}
      

References

Papers:

Blog post:

Code:

  • SBI a python library (PyTorch) that implements ABS, SNLE, SNPE, SNRE

Links to this note