Layer-Wise Training Convolutional Neural Networks With Smaller Filters for Human Activity Recognition Using Wearable Sensors

Recently, convolutional neural networks (CNNs) have set latest state-of-the-art on various human activity recognition (HAR) datasets. However, deep CNNs often require more computing resources, which limits their applications in embedded HAR. Although many successful methods been proposed to reduce memory and FLOPs of CNNs, they involve special network architectures designed for visual tasks, are not suitable HAR tasks with time series sensor signals, due remarkable discrepancy. Therefore, it is necessary develop lightweight models perform As filter the basic unit constructing deserves further research whether re-designing smaller filters applicable In paper, inspired by idea, we a CNN using Lego A lower-dimensional used as bricks be stacked conventional filters, does rely any structure. The local loss function train model. To our knowledge, this first paper that proposes ubiquitous wearable arena. experiment results five public datasets, UCI-HAR dataset, OPPORTUNITY UNIMIB-SHAR PAMAP2 WISDM dataset collected from either smartphones or multiple nodes, indicate novel can greatly computation cost over CNN, while achieving higher accuracy. That say, model smaller, faster accurate. Finally, evaluate actual performance an Android smartphone.