GenPark Elastic Weight Consolidation EWC Skill

Elastic Weight Consolidation (EWC) engine calculating Fisher Information Matrix diagonal penalties to mitigate catastrophic forgetting in continual learning.

Read more at GenPark and the GenPark MCP Catalog.

graph TD
    A[Task A Loss Minimization] --> B[Task A Optimal Weights theta_A*]
    B --> C[Compute Fisher Diagonal F_i]
    C --> D[Task B Training]
    D --> E[L_total = L_B + (lambda/2) * sum F_i * (theta - theta_A*)^2]
    E --> F[High Performance on Both Task A and Task B]

Loading

Features

• Diagonal Fisher Information Matrix (FIM) estimation.
• Quadratic elasticity penalty and analytical gradient calculation.
• Pure Python 3.9+ standard library.