slimgroup / ASPIRE.jl

ASPIRE: Amortized posteriors with Summaries that are Physics-based and Iteratively REfined

Code supporting "ASPIRE: Iterative Amortized Posterior Inference for Bayesian Inverse Problems" https://arxiv.org/abs/2405.05398

WARNING CODE AS IS: code is not in working condition. Please reach out to me for questions. I will be slowly bring the code base to running status.

In the meantime, the following benchmarks can be run using pregenerated datasets.

Benchmarks

Solving the non-linear high-dimensional wave-based inverse problem of Transcranial Ultrasound Computed Tomography (TUCT).

• Benchmark 1: Traditional SBI/likelihood-free-inference/amortized-posterior-inference.

Dataset

Download the brain prior samples paired with synthetic observationshere.

Goal Train an amortized posterior sampler of brain samples given raw observations.

Metrics Train on the first 1000 samples and compare with results in paper (Figure 7) on an unseen observation.

Code installation

To reproduce this project, do the following:

0. Download this code base. Notice that raw data needs to be downloaded independently.
1. Open a Julia console and do:

   julia> using Pkg
   julia> Pkg.add("DrWatson") # install globally, for using `quickactivate`
   julia> Pkg.activate("path/to/this/project")
   julia> Pkg.instantiate()
   

This will install all necessary packages for you to be able to run the scripts.