TimeCaps: Capturing Time Series Data With Capsule Networks
This work addresses the challenge of temporal understanding in signal processing for applications like healthcare and audio analysis, though it is incremental as it extends existing capsule network methods to a new domain.
The paper tackled the problem of capturing temporal relationships in 1D time series data by adapting capsule networks, achieving 96.21% accuracy on ECG signal beat classification and competitive results on audio command recognition.
Capsule networks excel in understanding spatial relationships in 2D data for vision related tasks. Even though they are not designed to capture 1D temporal relationships, with TimeCaps we demonstrate that given the ability, capsule networks excel in understanding temporal relationships. To this end, we generate capsules along the temporal and channel dimensions creating two temporal feature detectors which learn contrasting relationships. TimeCaps surpasses the state-of-the-art results by achieving 96.21% accuracy on identifying 13 Electrocardiogram (ECG) signal beat categories, while achieving on-par results on identifying 30 classes of short audio commands. Further, the instantiation parameters inherently learnt by the capsule networks allow us to completely parameterize 1D signals which opens various possibilities in signal processing.