Addition of ICNN Initialization Schemes #90
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
bunnech
added
documentation
marcocuturi
reviewed
Jun 30, 2022
marcocuturi
reviewed
Jun 30, 2022
marcocuturi
reviewed
Jun 30, 2022
marcocuturi
reviewed
Jun 30, 2022
marcocuturi
reviewed
Jun 30, 2022
marcocuturi
reviewed
Jun 30, 2022
michalk8
reviewed
Jun 30, 2022
| @@ -0,0 +1,520 @@ | |||
| { | |||
There was a problem hiding this comment.
Unnecessary braces in (dataloader_source, ..., )
In general, would prefer code cells to be formatted (black is ok if @marcocuturi is ok with it; wasn't able to find yapf-based formatter for notebooks).
Reply via ReviewNB
| @@ -0,0 +1,520 @@ | |||
| { | |||
There was a problem hiding this comment.
michalk8
pushed a commit
that referenced
this pull request
Jun 27, 2024
* Add ICNN formulation with new initialization schemes. * Adapt ICNN test. * Add notebook comparing both ICNN initialization schemes. * Update notebook on neural dual. * Add notebook to documentation. * Integration of comments by Marco. * Integration of comments by Marco. * Integration of comments by Marco. * Integration of comments by Marco. * 😵 * 🤯
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In this pull request, we add input convex neural network (ICNN) initialization schemes proposed in Bunne+(2022). Given a source and target measure, one aims at learning a map T between source and target parameterized via the gradient of an ICNN. To facilitate training, Bunne+(2022) propose to initialize the ICNN via different initialization schemes:
(i.) The
identityfunction, i.e., before training (and without a pretraining requirement), the gradient of the ICNN maps its input measure onto itself.(ii.) The
gaussianapproximation initializes the ICNN in such a way that the gradient of the ICNN maps a measure onto the Gaussian approximation of the opposite measure (utilizing closed-form results for Gaussian optimal transport maps).This pull request also updates the existing
NeuralDualto operate on the new ICNN initialization and a notebook comparing both initialization schemes.