An MLP initialized with weights derived analytically from 30 metric exemplars retrains faster and retains higher MNIST test accuracy at every training-set size from 60k down to 20k images compared with the same MLP started from random weights.
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 comparative analysis of a neural network with calculated weights and a neural network with random generation of weights based on the training dataset size
An MLP initialized with weights derived analytically from 30 metric exemplars retrains faster and retains higher MNIST test accuracy at every training-set size from 60k down to 20k images compared with the same MLP started from random weights.