Furthermore, weprovethat, under mild assumptions, a slight variation of WQL enjoys desirable theoretical properties in the tabular setting.

When n is independent ofδ our approach yields an algorithm whose sample complexityconvergesto 2n2 log 1δ asngrows.

Combined, these features provide both theoretical and practical simplicity.

That is, we propose to formulate continual learning as a sequence modeling problem, allowing advanced sequence models to be utilized for continual learning.