Title: Buzz, Choose, Forget: A Meta-Bandit Framework for Bee-Like Decision Making

Abstract: This study presents MAYA, a novel sequential imitation learning framework rooted in multi-armed bandit theory, which aims to replicate and forecast the decision-making processes of individual bees during contextualized foraging activities. The model incorporates the constraint of limited avian memory by utilizing a temporal window denoted as $\tau$. Analysis indicates that the optimal size for this window is approximately seven trials, with minor variations influenced by weather patterns. Evaluations conducted on real-world, simulated, and supplementary datasets (including mice) demonstrate that MAYA, especially when employing the Wasserstein distance metric, surpasses both classical statistical models and imitation-based baselines. Furthermore, the approach offers interpretability regarding individual learning strategies and facilitates the generation of realistic behavioral trajectories, thereby supporting future ecological applications.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC