Oja’s rule

Table of Contents

Hebb’s rule, but normalize weight after each increment

1. normalization

In here, defined as ensure the property that \(||w^t|| = 1\).

2. where to normalize

  1. after weight initializatoin
  2. after each weight increment

3. when to stop

convergence criteria
the largest change in weight smaller than a parameter \(\delta\) (small, positive, real), so \(max_i|w_i^t - w_i^{t-1} |\leq \delta\)
(no term)
check after weight increment + normalization
(no term)
note: the queried change here is on normalized weights, not raw \(\Delta w_i\)

Backlinks

Author: Linfeng He

Created: 2024-04-03 Wed 23:17