<ul><li>Contrastive learning has become popular for unsupervised representation learning, but optimal augmentation strategies for time series classification are not well-studied.</li><li>Existing time-domain augmentation methods are not specific to time series, leading to potential distortion of data with mismatched patterns.</li><li>A new perspective from the frequency domain is introduced, proposing FreRA for time series contrastive learning on classification tasks.</li><li>FreRA separates critical and unimportant frequency components, enhancing contrastive representation learning and generalization across diverse datasets.</li></ul>

FreRA: A Frequency-Refined Augmentation for Contrastive Learning on Time Series Classification

Discover more