ProSAR: Prototype-Guided Semantic Augmentation and Refinement for Time Series Contrastive Learning

Contrastive learning has advanced the representation learning across domains, yet its success relies on data augmentations that preserve semantic contents while providing the view diversities. Multivariate time series, however, are inherently noisy, non-stationary, and lack such intuitive semantic cues. Consequently, standard heuristic augmentations that ignore semantic parts may risk destroying critical temporal dependencies. Though some recent approaches attempt to isolate informative components, they typically rely on an implicit neural mechanism to infer semantics, thus limiting the interpretability and controllability. To address this, we propose ProSAR, an information-theoretic framework that leverages the explicit prototype alignment to guide semantic augmentations, and establish a feedback loop between the augmentation, contrastive learning, and prototype updates. Specifically, grounded in our proposed Prototype-Conditioned Information Bottleneck principle, we leverage the time-domain prototypes as explicit anchors to localize semantic segments, and develop a time–frequency augmentation strategy that retains prototype-consistent information while discarding noise. To promote semantically consistent prototypes for a reliable view generation, we design a dual-prototype loop where the augmented views are encoded into representations and then the learned representations are clustered to update latent prototypes, whose decoded feedback refines the time-domain prototypes for the next round of augmentation. Experiments on diverse time-series benchmarks demonstrate that ProSAR outperforms the other contrastive learning methods on downstream forecasting and classification tasks.

Many devices and systems record measurements over time, such as patient monitors, factory sensors, financial platforms, and wearable devices. To make sense of these data, machine learning models often need many labeled examples, but such labels can be costly or difficult to obtain. A common alternative is to train models by creating modified copies of the same signal and teaching the model that these copies still describe the same event. However, for time-series data, careless modifications may accidentally remove the very patterns that matter. We introduce ProSAR, a method that modifies time-series data more carefully. Instead of changing every part of a signal in the same way, ProSAR first learns representative patterns from the data and uses them as guides to find which parts of a signal are likely to be important. It then preserves and gently adjusts these important parts while reducing less useful noise. During training, these guides are also updated, so the method gradually becomes better at identifying meaningful patterns. Our experiments show that ProSAR helps models learn more useful time-series patterns, improving their ability to predict future values and recognize different types of signals, especially when labeled data are limited.