Avoid Touching Your Face: A Hand-to-face 3D Motion Dataset (COVID-away) and Trained Models for Smartwatches

Proceedings of the International Conference on Internet-of-Things Design and Implementation

Wahid

et al. 2021

Self Cite

In the Industry 4.0 era, Microcontrollers (MCUs) based tiny embedded sensor systems have become the sensing paradigm to interact with the physical world. In 2020, 25.6 billion MCUs were shipped, and over 250 billion MCUs are already operating in the wild. Such low-power, low-cost MCUs are being used as the brain to control diverse applications and soon will become the global digital nervous system. In an Industrial IoT setup, such tiny MCU-based embedded systems are equipped with anomaly detection models and mounted on production plant machines for monitoring the machine's health/condition. These models process the machine's health data (from temperature, RPM, vibration sensors) and raise timely alerts when it predicts/detects data patterns that show deviations from the normal operation state.In this demo, we train One Class Support Vector Machines (OC-SVM) based anomaly detection models and port the trained models to their MCU executable versions. We then deploy and execute the ported models on 4 popular MCUs and report their on-board inference performance along with their memory (Flash and SRAM) consumption. The steps/procedure that we show in the demo is generic, and the viewers can use it to efficiently port a wide variety of datasets-trained classifiers and execute them on different resource-constrained MCU and small CPU-based devices.

Section: Methodsmentioning

confidence: 99%

Porting and Execution of Anomalies Detection Models on Embedded Systems in IoT

Proceedings of the International Conference on Internet-of-Things Design and Implementation

Wahid

et al. 2021

Self Cite

“…To avoid this, we approached this problem using one-class learning approaches, as we had used in other recent work [32]. To produce E2G One-Class Classification (OCC) models, we trained only using benign data since, as briefly described earlier, it is not feasible to track hundreds of new malware forms and to collect their attack traffic data by infecting and observing the device.…”

Section: E2g Models Design and Evaluationmentioning

confidence: 99%

Edge2Guard: Botnet Attacks Detecting Offline Models for Resource-Constrained IoT Devices

Sundaram

2021 IEEE International Conference on Pervasive Computing and Communications Workshops and Other Affiliated Events (PerCom Work

et al. 2021

Self Cite

In today's IoT smart environments, dozens of MCUbased connected device types exist such as HVAC controllers, smart meters, smoke detectors, etc. The security conditions for these essential IoT devices remain unsatisfactory since: (i) many of them are built with cost as the driving design tenet, resulting in poor configurations and open design; (ii) their memory and computational resource constraints make it highly challenging to implement practical attack protection mechanisms; and (iii) currently, manufacturers use simplified light protocol versions to save memory for extra features (to boost sales). When such issues and vulnerabilities are exploited, devices can be compromised and converted into bots whereby severe DDoS attacks can be launched by a botmaster. Such tiny devices are safe only when connected to networks with defense mechanisms installed in their networking devices like routers and switches, which might not be present everywhere, e.g. on public/free Wi-Fi networks. To safeguard tiny IoT devices from cyberattacks, we provide resource-friendly standalone attack detection models termed Edge2Guard (E2G) that enable MCU-based IoT devices to instantly detect IoT attacks without depending on networks or any external protection mechanisms. During evaluation, our top-performing E2G models detected and classified ten types of Mirai and Bashlite malware with close to 100% detection rates.

“…In [8], authors present a generic pipeline named 'RCE-NN', that can fit, deploy, and execute a broad spectrum of CNN-based models on tiny IoT devices. For example, 'RCE-NN' has been used to run 'COVID-Away' models [9] on smartwatches, as well as DNNs trained for biometric authentication [10] on Alexa smart speakers.…”

Section: Ultra-fast Machine Learning Classifier Execution Onmentioning

confidence: 99%

Ultra-fast Machine Learning Classifier Execution on IoT Devices without SRAM Consumption

2021 IEEE International Conference on Pervasive Computing and Communications Workshops and Other Affiliated Events (PerCom Work

Breslin

et al. 2021

Self Cite

With the introduction of edge analytics, IoT devices are becoming smart and ready for AI applications. A few modern ML frameworks are focusing on the generation of small-size ML models (often in kBs) that can directly be flashed and executed on tiny IoT devices, particularly the embedded systems. Edge analytics eliminates expensive device-to-cloud communications, thereby producing intelligent devices that can perform energyefficient real-time offline analytics. Any increase in the training data results in a linear increase in the size and space complexity of the trained ML models, making them unable to be deployed on IoT devices with limited memory. To alleviate the memory issue, a few studies have focused on optimizing and fine-tuning existing ML algorithms to reduce their complexity and size. However, such optimization is usually dependent on the nature of IoT data being trained. In this paper, we presented an approach that protects model quality without requiring any alteration to the existing ML algorithms. We propose SRAM-optimized implementation and efficient deployment of widely used standard/stable ML-frameworks classifier versions (e.g., from Python scikit-learn). Our initial evaluation results have demonstrated that ours is the most resource-friendly approach, having a very limited memory footprint while executing large and complex ML models on MCU-based IoT devices, and can perform ultra-fast classifications while consuming 0 bytes of SRAM. When we tested our approach by executing it on a variety of MCU-based devices, the majority of models ported and executed produced 1-4x times faster inference results in comparison with the models ported by the sklearn-porter, m2cgen, and emlearn libraries.