https://github.com/avik-pal/localregneuralde.jl
Local regularization of Neural Differential Equations
https://github.com/avik-pal/localregneuralde.jl
Last synced: about 1 month ago
JSON representation
Local regularization of Neural Differential Equations
- Host: GitHub
- URL: https://github.com/avik-pal/localregneuralde.jl
- Owner: avik-pal
- License: mit
- Created: 2022-09-12T16:01:13.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2023-10-29T00:41:31.000Z (over 1 year ago)
- Last Synced: 2025-01-21T04:41:57.932Z (3 months ago)
- Language: Julia
- Size: 67.4 KB
- Stars: 7
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# LocalRegNeuralDE: Some Black Boxes Were Meant to Remain Closed!
Official Implementation of ICML 2023 Paper [Locally Regularized Neural Differential Equations: Some Black Boxes Were Meant to Remain Closed!](https://arxiv.org/pdf/2303.02262.pdf)
Extension of [RegNeuralDE](https://arxiv.org/abs/2105.03918) to allow for local
regularization.
This is a research repository. Most users should wait for the functionality to be available
through packages like `DiffEqFlux.jl`.
## Summary
* (Randomized) Local Regularization of Neural ODEs yield similar trajectories to global
regularization.
* Local Regularization is simpler to implement since it doesn’t require
discretize-then-optimize. Rather it can utilize the more commonly used
optimize-then-discretize approach.
* Additionally introduces benefits for cases where the “blackbox cannot be opened” entirely
(like in diffusion models)
## Installation
This package is not registered in General Registry. It can be installed using:
```julia
using Pkg
pkg"add https://github.com/avik-pal/LocalRegNeuralDE.jl"
```