The gradient flow of linear convolutional networks is a Riemannian gradient flow on function space independent of initialization for D-dimensional convolutions with D at least 2 or with stride greater than 1 in 1D.
Abide by the law and follow the flow: Conservation laws for gradient flows
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
-
The Riemannian Geometry Associated to Gradient Flows of Linear Convolutional Networks
The gradient flow of linear convolutional networks is a Riemannian gradient flow on function space independent of initialization for D-dimensional convolutions with D at least 2 or with stride greater than 1 in 1D.