“…Any standalone MLC classifier such as MLSAkNN (Roseberry et al, 2021), Multilabel Hoeffding Tree (Read, Bifet, Holmes, & Pfahringer, 2012), or i‐SOUP Tree (Osojnik et al, 2017) could be suitable base learners in this context. To extend OMS‐MAB to delayed label setting, we borrow from (Hu & Kantardzic, 2017) and propose keeping a fixed‐size reservoir of labeled samples. The reservoir is updated with incoming data based on instance and label arrival times, while discarding excessively delayed samples which potentially represent obsolete concepts.…”