Augmentation Policy

Augmentation policy research focuses on automatically designing optimal data augmentation strategies for machine learning models, aiming to improve model generalization and performance without manual intervention. Current research explores various approaches, including evolutionary algorithms, Bayesian optimization, and reinforcement learning, to learn augmentation policies tailored to specific tasks and datasets, often employing bi-level optimization or contextual bandit frameworks. These advancements have significant implications for diverse fields, enhancing the efficiency and robustness of models in areas such as image classification, time-series forecasting, and even optimizing the performance of exact solvers for complex problems.

Papers