Are there suitable machine learning techniques that may be applied to a continual stream of data and update its models for data that it believes to be different to the most representative case but still correct?
If so, under what scenarios would some be particularly useful?
