A Near-Optimal Best-of-Both-Worlds Algorithm for Online Learning with Feedback Graphs
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Documents
- Fulltext
Submitted manuscript, 336 KB, PDF document
We consider online learning with feedback graphs, a sequential decision-making framework where the learner's feedback is determined by a directed graph over the action set. We present a computationally efficient algorithm for learning in this framework that simultaneously achieves near-optimal regret bounds in both stochastic and adversarial environments. The bound against oblivious adversaries is O~(αT−−−√), where T is the time horizon and α is the independence number of the feedback graph. The bound against stochastic environments is O((lnT)2maxS∈I(G)∑i∈SΔ−1i) where I(G) is the family of all independent sets in a suitably defined undirected version of the graph and Δi are the suboptimality gaps. The algorithm combines ideas from the EXP3++ algorithm for stochastic and adversarial bandits and the EXP3.G algorithm for feedback graphs with a novel exploration scheme. The scheme, which exploits the structure of the graph to reduce exploration, is key to obtain best-of-both-worlds guarantees with feedback graphs. We also extend our algorithm and results to a setting where the feedback graphs are allowed to change over time.
Original language | English |
---|---|
Title of host publication | Advances in Neural Information Processing Systems 35 (NeurIPS 2022) |
Publisher | NeurIPS Proceedings |
Publication date | 2022 |
Pages | 35035-35048 |
ISBN (Electronic) | 9781713871088 |
Publication status | Published - 2022 |
Event | 36th Conference on Neural Information Processing Systems (NeurIPS 2022). - New Orleans/ Virtual, United States Duration: 28 Nov 2022 → 9 Dec 2022 |
Conference
Conference | 36th Conference on Neural Information Processing Systems (NeurIPS 2022). |
---|---|
Land | United States |
By | New Orleans/ Virtual |
Periode | 28/11/2022 → 09/12/2022 |
ID: 383100739