Official Implementation of ICML 2023 Paper Locally Regularized Neural Differential Equations: Some Black Boxes Were Meant to Remain Closed!
Extension of RegNeuralDE 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
.
-
(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)
This package is not registered in General Registry. It can be installed using:
using Pkg
pkg"add https://github.com/avik-pal/LocalRegNeuralDE.jl"