ReAugment: Targeted Few-Shot Time Series Augmentation via Model Zoo-Guided Reinforcement Learning

Few-shot time series forecasting is fundamentally challenged by the scarcity of high-quality training data and the risk of severe overfitting. To address this issue, we propose ReAugment, a reinforcement learning (RL) framework that explicitly learns where and how to augment time series data. ReAugment maintains a zoo of forecasting models and measures prediction diversity across them to identify training samples that are most prone to overfitting. These samples serve as anchor points and are used as inputs to the data augmentation process. We then employ an RL approach to learn transformation policies, using a model zoo-guided reward function to bias the transformed data to overfit-prone regions of the training distribution that are most beneficial for generalization. A key advantage of the RL formulation is that it avoids backpropagating gradients through the forecasting models, thereby mitigating gradient vanishing. Experiments across diverse forecasting architectures demonstrate the effectiveness of ReAugment in both few-shot and standard time series forecasting.

Time series forecasting models often overfit when only limited training data is available. ReAugment improves few-shot forecasting by learning where and how to augment data. It first builds a small zoo of forecasting models and uses their prediction disagreement to identify overfit-prone training samples. These samples are then used as anchors for a variational generative model, whose prior network is further optimized with reinforcement learning. The reward encourages generated samples to be both realistic and useful for reducing model disagreement. In this way, ReAugment creates targeted synthetic data around the most fragile regions of the training distribution, leading to more robust forecasting models.