When Exploration Comes for Free with Mixture-Greedy: Do we need UCB in Diversity-Aware Multi-Armed Bandits?

TL;DR

This paper challenges the necessity of UCB exploration bonuses in diversity-aware multi-armed bandits for generative AI, showing that a simple greedy approach often outperforms UCB-based methods due to implicit exploration.

Contribution

The paper demonstrates that in diversity-aware model selection, implicit exploration from the objective geometry can replace explicit UCB bonuses, leading to faster convergence and better performance.

Findings

Mixture-Greedy converges faster than UCB-based methods.

Implicit exploration arises from diversity-aware objectives.

Theoretical analysis explains conditions for implicit exploration.

Abstract

Efficient selection among multiple generative models is increasingly important in modern generative AI, where sampling from suboptimal models is costly. This problem can be formulated as a multi-armed bandit task. Under diversity-aware evaluation metrics, a non-degenerate mixture of generators can outperform any individual model, distinguishing this setting from classical best-arm identification. Prior approaches therefore incorporate an Upper Confidence Bound (UCB) exploration bonus into the mixture objective. However, across multiple datasets and evaluation metrics, we observe that the UCB term consistently slows convergence and often reduces sample efficiency. In contrast, a simple \emph{Mixture-Greedy} strategy without explicit UCB-type optimism converges faster and achieves even better performance, particularly for widely used metrics such as FID and Vendi where tight confidence bounds are difficult to construct. We provide theoretical insight explaining this behavior: under transparent structural conditions, diversity-aware objectives induce implicit exploration by favoring interior mixtures, leading to linear sampling of all arms and sublinear regret guarantees for entropy-based, kernel-based, and FID-type objectives. These results suggest that in diversity-aware multi-armed bandits for generative model selection, exploration can arise intrinsically from the objective geometry, questioning the necessity of explicit confidence bonuses.