Transitioning Continual Learning from Parameters to Systems

Continual learning is shifting from parameter-centric weight adaptation toward system-level adaptation. This includes on-policy learning, test-time training, and the use of external components like memory and skill libraries to extend model capabilities.