Abstract: Wearable human activity recognition (HAR) has greatly benefited from deep convolutional neural networks. However, while increasing network width or depth improves accuracy, it also leads to ...