Joint-Label Learning by Dual Augmentation for Time Series Classification

Recently, deep neural networks (DNNs) have achieved excellent performance on time series classification. However, DNNs require large amounts of labeled data for supervised training. Although augmentation can alleviate this problem, the standard approach assigns same label to all augmented samples from source. This leads expansion distribution such that classification boundaries may be even harder determine. In paper, we propose Joint-label learning by Dual Augmentation (JobDA), which enrich training without expanding original data. Instead, apply simple transformations and give these modified new labels, so model has distinguish between data, as well separating classes. sharpens around series, results in superior performance. We use Time Series Warping our transformations: shrink stretch different regions like a fun-house mirror. Experiments conducted extensive time-series datasets show JobDA improve small datasets. Moreover, verify better generalization ability compared with conventional augmentation, visualization analysis further demonstrates learn more compact clusters.