Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/joerihermans/amortized-experimental-design

A demonstration on how to enable active sciencing by amortizing Bayesian experimental design.
https://github.com/joerihermans/amortized-experimental-design

Last synced: 7 days ago
JSON representation

A demonstration on how to enable active sciencing by amortizing Bayesian experimental design.

Host: GitHub
URL: https://github.com/joerihermans/amortized-experimental-design
Owner: JoeriHermans
License: bsd-3-clause
Created: 2021-09-12T09:15:19.000Z (over 3 years ago)
Default Branch: master
Last Pushed: 2021-09-12T10:19:52.000Z (over 3 years ago)
Last Synced: 2024-10-28T21:49:03.678Z (about 2 months ago)
Language: Jupyter Notebook
Size: 5.49 MB
Stars: 0
Watchers: 1
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md
- License: LICENSE

Awesome Lists containing this project

README

[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/JoeriHermans/amortized-experimental-design/master)

## The idea

Which **experimental configuration** $\psi$ yields the largest expected gain in information?
The utility $U(\psi)$ can be expressed as the exected reduction in entropy between the prior and the posterior, i.e.,

$U(\psi) = \mathbb{E}\left[ \mathbb{H}\left[p(\vartheta)\right] - \mathbb{H}\left[p(\vartheta\vert x,\psi)\right] \right] \propto \mathbb{E}_{p(\vartheta,x\vert\psi)}\left[\log\frac{p(\vartheta\vert x,\psi)}{p(\vartheta)}\right]$ .

This quantity is especially challenging because the posterior is intractable in most practical applications. We can however, draw samples from the likelihood model (our simulator).
We seek to obtain the experimental configuration which maximizes the utility: $\psi^* = \text{argmax}_\psi U(\psi)$

**Problem**: for every evaluation of the utility, the simulator needs to be called because the
expectation depends on $p(\vartheta,x\vert\psi)$ . Slow and cumbersome!

**Proposal:** Reweigh the samples from the joint $p(\vartheta,x,\psi)$ to approximate $\mathbb{E}_{p(\vartheta,x\vert\psi)}\left[\log\frac{p(\vartheta\vert x,\psi)}{p(\vartheta)}\right]$ with several ratio estimators that can be trained on samples from the joint alone!
In doing so, we can estimate the EIG for every experimental configuration by reusing presimulated samples for specific experimental configurations. In addition, this specification allows us to maximize the expected information gain through gradient ascent by simply backpropagating through the neural networks. For more details check the notebook!

## License

See LICENSE file.