Optimal Multiclass Online Regret with Arbitrary Label Spaces
What it is
This paper determines the optimal regret for agnostic multiclass online learning with arbitrary or infinite label spaces. It proves a regret upper bound of O(min{T, sqrt{dT}}) for hypothesis classes with finite Littlestone dimension d, matching known lower bounds and resolving a question posed by Hanneke et al. (2023).
The problem it solves
The lack of a proven optimal regret bound for agnostic multiclass online learning with arbitrary label spaces, which previously left a gap between known upper and lower bounds.
What you could use it for
- Designing online learning algorithms for multiclass classification with large or unbounded label sets
- Theoretical analysis of regret bounds in non-parametric online learning
- Extending binary online learning results to multiclass settings
Why it matters
Establishes the theoretical limits of performance for multiclass online learning algorithms, providing a definitive benchmark for algorithm design in scenarios with complex label spaces.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.