We study multitask online learning in a setting where agents can only exchange information with their neighbors on a given arbitrary communication network. We introduce MT-CO2OL, a decentralized algorithm for this setting whose regret depends on the interplay between the task similarities and the network structure. Our analysis shows that the regret of MT-CO2OL is never worse (up to constants) than the bound obtained when agents do not share information. On the other hand, our bounds significantly improve when neighboring agents operate on similar tasks. In addition, we prove that our algorithm can be made differentially private with a negligible impact on the regret. Finally, we provide experimental support for our theory.
Multitask Online Learning: Listen to the Neighborhood Buzz / J. Achddou, N. Cesa Bianchi, P. Laforgue (PROCEEDINGS OF MACHINE LEARNING RESEARCH). - In: AISTATS / [a cura di] S. Dasgupta, S. Mandt, Y. Li. - [s.l] : ML Research Press, 2024. - pp. 1846-1854 (( Intervento presentato al 27. convegno International Conference on Artificial Intelligence and Statistics : 2nd through 4th May tenutosi a Valencia nel 2024.
Multitask Online Learning: Listen to the Neighborhood Buzz
J. AchddouPrimo
;N. Cesa BianchiPenultimo
;P. LaforgueUltimo
2024
Abstract
We study multitask online learning in a setting where agents can only exchange information with their neighbors on a given arbitrary communication network. We introduce MT-CO2OL, a decentralized algorithm for this setting whose regret depends on the interplay between the task similarities and the network structure. Our analysis shows that the regret of MT-CO2OL is never worse (up to constants) than the bound obtained when agents do not share information. On the other hand, our bounds significantly improve when neighboring agents operate on similar tasks. In addition, we prove that our algorithm can be made differentially private with a negligible impact on the regret. Finally, we provide experimental support for our theory.File | Dimensione | Formato | |
---|---|---|---|
achddou24a.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
7.32 MB
Formato
Adobe PDF
|
7.32 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.