I would argue there is no such thing. The model will after the update now incooperate your traning data as a seen example, clever use of optimization would enab

I would argue there is no such thing. The model will after the update now incooperate your traning data as a seen example, clever use of optimization would enable you to partly reconstruct the example.