English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Human Activity Recognition (HAR) based on inertial data is an increasinglydiffused task on embedded devices, from smartphones to ultra low-power sensors.Due to the high computational complexity of deep learning models, most embeddedHAR systems are based on simple and not-so-accurate classic machine learningalgorithms. This work bridges the gap between on-device HAR and deep learning,proposing a set of efficient one-dimensional Convolutional Neural Networks(CNNs) deployable on general purpose microcontrollers (MCUs). Our CNNs areobtained combining hyper-parameters optimization with sub-byte andmixed-precision quantization, to find good trade-offs between classificationresults and memory occupation. Moreover, we also leverage adaptive inference asan orthogonal optimization to tune the inference complexity at runtime based onthe processed input, hence producing a more flexible HAR system. Withexperiments on four datasets, and targeting an ultra-low-power RISC-V MCU, weshow that (i) We are able to obtain a rich set of Pareto-optimal CNNs for HAR,spanning more than 1 order of magnitude in terms of memory, latency and energyconsumption; (ii) Thanks to adaptive inference, we can derive >20 runtimeoperating modes starting from a single CNN, differing by up to 10% inclassification scores and by more than 3x in inference complexity, with alimited memory overhead; (iii) on three of the four benchmarks, we outperformall previous deep learning methods, reducing the memory occupation by more than100x. The few methods that obtain better performance (both shallow and deep)are not compatible with MCU deployment. (iv) All our CNNs are compatible withreal-time on-device HAR with an inference latency <16ms. Their memoryoccupation varies in 0.05-23.17 kB, and their energy consumption in 0.005 and61.59 uJ, allowing years of continuous operation on a small battery supply.

Human Activity Recognition on Microcontrollers with Quantized and Adaptive Deep Neural Networks