Login / Signup
Stochastic Collapse: How Gradient Noise Attracts SGD Dynamics Towards Simpler Subnetworks.
Feng Chen
Daniel Kunin
Atsushi Yamamura
Surya Ganguli
Published in:
NeurIPS (2023)
Keyphrases
</>
gradient estimation
noise reduction
noise level
monte carlo
dynamical systems
image noise
additive noise
random noise
missing data
signal to noise ratio
gradient information
gradient direction
cross validation
network structure
dynamic model
image structure