On the Generalization Benefit of Noise in Stochastic Gradient Descent.

Samuel L. Smith Erich Elsen Soham De

Published in: CoRR (2020)

Keyphrases

stochastic gradient descent
least squares
loss function
matrix factorization
step size
random forests
noise level
regularization parameter
weight vector
noise reduction
multiple kernel learning
semi supervised
prediction accuracy
arbitrary shape
online algorithms