The forward pass of a neural network is an Euler step of a relaxation PDE: the activation function is the local equilibrium and the weights act as advection-diffusion coefficients.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A note on the physical interpretation of neural PDE's
The forward pass of a neural network is an Euler step of a relaxation PDE: the activation function is the local equilibrium and the weights act as advection-diffusion coefficients.