Distributed Adversarial Bandits
Event details
| Date | 20.03.2026 |
| Hour | 11:00 › 12:00 |
| Speaker | Prof. Nicolò Cesa-Bianchi, Computer Science at Università degli Studi di Milano, Italy |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract:
We study distributed adversarial bandits, where N agents cooperate to minimize the global average loss while observing only their own local losses. We show that the minimax regret for this problem is expressed as a combination of a term accounting for the communication cost and a term accounting for the bandit feedback cost. Our upper bound, which significantly improves over the previous best bound, is based on a novel black-box reduction to bandits with delayed feedback which can be extended to more general settings, like distributed linear bandits. In the talk, we will place this result in the context of the literature on multi-agent online learning algorithms.
Joint work with Hao Qiu and Mengxiao Zhang
Biography:
Nicolò Cesa-Bianchi is professor of Computer Science at Università degli Studi di Milano. His main research interests are the design and analysis of algorithms for machine learning and sequential decision-making. He is co-author of the monographs "Prediction, Learning, and Games" and "Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems". He served as President of the Association for Computational Learning and co-chaired the program committees of some of the most important machine learning conferences, including NeurIPS and COLT. He is the recipient of a Google Research Award, a Xerox Foundation Award, a Criteo Faculty Award, a Google Focused Award, an IBM Research Award, and the FOCS Test-of-Time Award. He is ELLIS fellow, member of the ELLIS board, and co-director of the ELLIS program on Interactive Learning and Interventional Representations. He is member of the Italian Academy of Sciences.
We study distributed adversarial bandits, where N agents cooperate to minimize the global average loss while observing only their own local losses. We show that the minimax regret for this problem is expressed as a combination of a term accounting for the communication cost and a term accounting for the bandit feedback cost. Our upper bound, which significantly improves over the previous best bound, is based on a novel black-box reduction to bandits with delayed feedback which can be extended to more general settings, like distributed linear bandits. In the talk, we will place this result in the context of the literature on multi-agent online learning algorithms.
Joint work with Hao Qiu and Mengxiao Zhang
Biography:
Nicolò Cesa-Bianchi is professor of Computer Science at Università degli Studi di Milano. His main research interests are the design and analysis of algorithms for machine learning and sequential decision-making. He is co-author of the monographs "Prediction, Learning, and Games" and "Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems". He served as President of the Association for Computational Learning and co-chaired the program committees of some of the most important machine learning conferences, including NeurIPS and COLT. He is the recipient of a Google Research Award, a Xerox Foundation Award, a Criteo Faculty Award, a Google Focused Award, an IBM Research Award, and the FOCS Test-of-Time Award. He is ELLIS fellow, member of the ELLIS board, and co-director of the ELLIS program on Interactive Learning and Interventional Representations. He is member of the Italian Academy of Sciences.
Practical information
- General public
- Free
Organizer
- Prof. Maryam Kamgarpour