SGD weight spectra of networks trained on stochastic processes are portfolio allocation spectra, and isotropic objective perturbations leave that spectrum invariant up to scale.
Jean-Philippe Bouchaud and Marc Mézard
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
CONDITIONAL 2representative citing papers
A proportional wealth tax preserves the Gini coefficient at finite times under homogeneous returns, while a progressive tax acts as a confining potential that reduces inequality at a rate determined by the progressivity parameter.
citing papers explorer
-
Spectral Portfolio Theory: From SGD Weight Matrices to Wealth Dynamics
SGD weight spectra of networks trained on stochastic processes are portfolio allocation spectra, and isotropic objective perturbations leave that spectrum invariant up to scale.
-
From Gravity to Confinement: Wealth Redistribution as Optimal Drift Design in the Fokker-Planck Framework
A proportional wealth tax preserves the Gini coefficient at finite times under homogeneous returns, while a progressive tax acts as a confining potential that reduces inequality at a rate determined by the progressivity parameter.