Continual Learning in Transition
The survey argues continual learning is moving from weight updates to system-level adaptation.
The authors frame the shift across three axes: when learning happens, how updates are made, and where adaptation lives. Their scope includes on-policy learning, test-time training, and external components such as memory, skill libraries, and interaction protocols. The paper positions these as expanding continual learning beyond static model parameters in the LLM and agentic-AI era. ArXiv · AI/CL/LG's note
The authors frame the shift across three axes: when learning happens, how updates are made, and where adaptation lives. Their scope includes on-policy learning, test-time training, and external components such as memory, skill libraries, and interaction protocols. The paper positions these as expanding continual learning beyond static model parameters in the LLM and agentic-AI era. ArXiv · AI/CL/LG's note
score 4