3 results
Filters
Settings
Now showing1 - 3 of 3
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , MiniLearn: on-device learning for low-power IoT devices(2022-10); ; Recent advances in machine learning enable new, intelligent applications in the Internet of Things. For example, today’s smartwatches use Deep Neural Networks (DNNs) to detect and classify human activities. The training of DNNs, however, is done offline with previously collected and labeled datasets using extensive computational resources such as GPUs on cloud services. Once being quantized and deployed on an IoT device, a DNN commonly remains unchanged. We argue that this static nature of trained DNNs strongly limits their flexibility to adapt to requirements that change dynamically. For example, the device may need to adjust on the fly to the limited memory and energy resources, but only the retraining or pruning of the DNN in the cloud can address these issues. Moreover, the user may need to add new classes or refine existing ones, due to different problem domains materializing dynamically. Retraining DNNs requires a high volume of data collected from IoT devices and transmitted to the cloud. However, IoT devices depend on energy-efficient communication with limited reliability and network bandwidth. In addition, cloud storage of extensive IoT data raises significant privacy concerns. This paper introduces MiniLearn that enables re-training of DNNs on resource-constrained IoT devices. MiniLearn allows IoT devices to re-train and optimize pre-trained, quantized neural networks using IoT data collected during deployment of an IoT device. We show that MiniLearn speeds up inference by a factor of up to 2 and requires up to 50% less memory compared to original DNN. In addition, MiniLearn increases classification accuracy for a sub-set by 3% to 9% of the original DNN.Publicationtype: Conference Paper57 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , TraceBand: privacy-preserving contact tracing on low-power wristbands(2022-10); TraceBand is a low-power and versatile wristband for contact tracing during pandemic outbreaks. Running on a 64 MHz Cortex-M4 microprocessor with 256 KB of memory, TraceBand targets a low-cost design. Still, it seamlessly integrates into the smartphone-based tracing, extending the reach and, therefore, the effectiveness of the Exposure Notification protocol by Google and Apple. As the wristband is restricted to Bluetooth Low Energy for tracing and communication, companion devices and gateways serve as synchronization points and allow the device to offload the recorded contacts for an off-device risk analysis or download the information of infected contacts to perform the risk analysis on the device itself. The presented design ensures compatibility from the ground up and tackles occurring challenges due to resource reduction. Evaluations cover the applicability and energy consumption of the Exposure Notification protocol on resource-constrained devices and the on-device identification of risk contacts. The results show that the contact exchange suits battery-driven, resource-constrained devices, as TraceBand requires 2 mAh per day. At the expanse of battery life, up to 5,000,000 risk keys can be checked on the device daily, increasing consumption to roughly 80 mAh for a single day.Publicationtype: Conference Paper25 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Equalization for high-bitrate acoustic backscatter communication in metals(2022-10); Acoustic backscatter communication allows sensors in challenging environments to transmit data with negligible energy cost and is especially useful where electromagnetic waves are infeasible. In acoustic channels, strong inter-symbol interference (ISI) poses a major challenge, which is usually—in conventional communications—mitigated with equalization. Backscatter channels, however, are nonlinear, and standard equalization techniques are not directly applicable. Our work demonstrates that, by mapping the tag’s load to emerging steady-states, the nonlinear channel can be split into a map and a linear channel, enabling standard equalization techniques. Employing our custom reader and tag hardware design in the guided-wave (GW) metal channel, we demonstrate that data rates can be increased more than five-fold compared to no equalization with negligible additional cost. We compare different algorithms for equalization, timing recovery, and interpolation. Moreover, we investigate the time-variability of the GW channel under realistic scenarios, e.g., varying temperatures and physical stress, and present an effective countermeasure to adapt to the time variance.Publicationtype: Conference Paper169