[2308.04620] Multiclass On-line Learnability beneath Bandit Suggestions

Obtain a PDF of the paper titled Multiclass On-line Learnability beneath Bandit Suggestions, by Ananth Raman and 4 different authors

Obtain PDF
HTML (experimental)

Summary:We examine on-line multiclass classification beneath bandit suggestions. We prolong the outcomes of Daniely and Helbertal [2013] by displaying that the finiteness of the Bandit Littlestone dimension is critical and adequate for bandit on-line learnability even when the label area is unbounded. Furthermore, we present that, in contrast to the full-information setting, sequential uniform convergence is critical however not adequate for bandit on-line learnability. Our consequence enhances the latest work by Hanneke, Moran, Raman, Subedi, and Tewari [2023] who present that the Littlestone dimension characterizes on-line multiclass learnability within the full-information setting even when the label area is unbounded.