A Good Start Matters: Enhancing Continual Learning with Data-Driven Weight Initialization