Signature representations that are extracted by convolutional neural networks (CNN) can achieve low error rates. However, a trade-off exists between such models' complexities and hand-crafted features' slightly higher error ...