no code implementations • 7 Dec 2023 • Christian Goodbrake
Natural gradient descent has a remarkable property that in the small learning rate limit, it displays an invariance with respect to network reparameterizations, leading to robust training behavior even for highly covariant network parameterizations.
no code implementations • 22 Aug 2022 • Christian Goodbrake, David Beers, Travis B. Thompson, Heather A. Harrington, Alain Goriely
This sequence of edges defines a filtration of the graph.