Fourier-diagonalized natural gradients coincide with Sobolev mirror descent when their spectral symbols match and are otherwise preconditioned by it, yielding a new Spectral Natural Gradient implementation.
Amari, Natural gradient works efficiently in learning, Neural Com- putation 10 (2) (1998) 251–276
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.NA 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Fourier-Diagonalized Natural Gradients and Sobolev Mirror Descent
Fourier-diagonalized natural gradients coincide with Sobolev mirror descent when their spectral symbols match and are otherwise preconditioned by it, yielding a new Spectral Natural Gradient implementation.