On the Computation of the Fisher Information in Continual Learning
This is an incremental clarification for researchers using EWC in continual learning.
The paper tackles the inconsistent computation of Fisher Information in Elastic Weight Consolidation for continual learning, empirically comparing implementations and suggesting that many reported results could be improved.
One of the most popular methods for continual learning with deep neural networks is Elastic Weight Consolidation (EWC), which involves computing the Fisher Information. The exact way in which the Fisher Information is computed is however rarely described, and multiple different implementations for it can be found online. This blog post discusses and empirically compares several often-used implementations, which highlights that many currently reported results for EWC could likely be improved by changing the way the Fisher Information is computed.