You may be interested in TITANS:

Test-Time Learning: The model updates its own memory weights while running an inference task.