The authors define and compute a Boltzmann entropy for the deep linear network via group-orbit foliations of the balanced manifold and show that observable Riemannian geometry arises by submersion from this manifold.
Brockett , Modeling the transient behavior of stochastic gradient algorithms , in 2011 50th IEEE Conference on Decision and Control and European Control Conference, IEEE, 2011, pp
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
An entropy formula for the Deep Linear Network
The authors define and compute a Boltzmann entropy for the deep linear network via group-orbit foliations of the balanced manifold and show that observable Riemannian geometry arises by submersion from this manifold.