Scaling differential equations via transferable dynamics. We focus on generalising dysnmical systems to new environments. We treat a each of those environments as a node, and the edges carry the difference in network weight norms between them.
- Faster discovery of dynamical systems.
- Better learning with advanced regularisation of the dynamics.
- Transfer learning with all its advantages (sim2real, etc...).
- Interpretable representations of high dimensional data via latent variable models.
pip install graphpint
- Build single examples script for neural ODE, APHYNITY, etc.
- Build a unifying framework
- JAX
- Equinox
- Jraph