This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Berrenur Saylam;
(2) Ozlem Durmaz ¨ ˙Incel. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Berrenur Saylam; (2) Ozlem Durmaz ¨ ˙Incel. Table of Links Highlights Highlights Abstract & Introduction Abstract & Introduction Federated Learning: Collaborative and Privacy-Preserving Model Training Federated Learning: Collaborative and Privacy-Preserving Model Training State of the Art Federated Learning Software Platforms and Testbeds State of the Art Federated Learning Software Platforms and Testbeds Sensors and Edge-Sensing Devices Sensors and Edge-Sensing Devices Federated Learning Applications on Sensing Devices Federated Learning Applications on Sensing Devices Discussion Discussion Conclusions, Acknowledgment, and References Conclusions, Acknowledgment, and References 4. Sensors and Edge-Sensing Devices In this study, we focus on edge sensing devices such as IoT sensors, mobile sensing devices (smartphones), wearables, and many other sensing devices that are used in the application domains of sensor-based edge devices: medical, environmental, automotive, industrial, aerospace, and agricultural applications as seen in Figure 4. The devices possess three fundamental abilities: context, computing, and connectivity Jaren et al. (2015). Context refers to measuring information using sensors while extracting knowledge from the sensed data. Connectivity enables communication with other devices or central servers to facilitate knowledge transfer Jaren et al. (2015). The emergence of sensor technology has revolutionized data collection and computation capabilities, giving rise to the concept of the Internet of Things (IoT) Want et al. (2015). These small-sized sensors find applications in various domains, including health monitoring (such as heart rate fluctuations and step counts), communication and environment sensing in autonomous vehicles (using cameras and proximity sensors) Rehman et al. (2017), as well as marketplaces, agriculture, smart manufacturing, smart buildings, and energy management McCann et al. (2018). Besides dedicated IoT devices, our smartphones and wearables, such as smartwatches, also have sensors. Wearable data analysis has gained significant popularity over the past decade, with Human Activity Recognition being a typical application area Feng et al. (2020); Sozinov et al. (2018); Ek et al. (2020). Researchers have started focusing on more complex activities and behaviours, such as smoking, emotion, and stress recognition, rather than solely concentrating on basic activity types like walking and running. Various sensors can be utilized in the target devices, including motion, electrodermal activity, blood volume pulse, heart rate variability (HRV), accelerometer, gyroscope, magnetometer, pressure, proximity, temperature, GPS, sound, image, water quality, chemical, gas, smoke, infrared, humidity, and optical sensors Srinivasan et al. (2019). Sensing devices’ hardware properties significantly impact their performance and capabilities. The central processing unit (CPU) is a critical component, that executes instructions and calculations, and its performance is determined by clock speed, core number, and cache memory. Comparing CPU specifications across devices like smartwatches, Raspberry Pi, etc., re veals differences in processing power and computational capabilities. For instance, smartwatches have low-power CPUs suitable for basic tasks like notifications and fitness tracking while Raspberry Pi devices, with quadcore ARM-based CPUs, offer better capabilities than smartwatches and can handle lightweight desktop applications and basic gaming. However, smart phones equipped with multi-core CPUs provide significantly higher processing power, making them ideal for demanding tasks like video editing, and machine learning, as well as running resource-intensive software efficiently. GPUs, specialized processors, enhance computational power and enable advanced graphical tasks in image processing, augmented reality, and machine learning. While some smartwatches and Raspberry Pi models may have integrated graphics capabilities as part of their SoC (System on a Chip), they are not typically equipped with dedicated GPUs like those found in PCs. Dedicated GPUs in PCs are specifically designed for high-performance graphical tasks, such as gaming, video rendering, and machine learning, providing much higher computational power for these tasks compared to integrated graphics found in smartwatches and Raspberry Pi devices. Memory is another crucial hardware component in all kinds of devices. Random access memory (RAM) provides temporary storage for data that the device uses during operation. Devices with larger RAM capacities can handle more extensive datasets and run multiple applications simultaneously without experiencing significant performance degradation. Sensing devices like smartwatches and Raspberry Pi have significantly lower memory capacity than a typical PC. Smart-watches and Raspberry Pi often have 512 MB to 1 GB RAM, while PCs typically have 8 GB to 16 GB or more. Smart-watches and Raspberry Pi rely on smaller internal storage (4 GB to 8 GB) or external MicroSD cards (16 GB to 128 GB) for data storage. In contrast, PCs have more extensive internal storage options (256 GB to 1 TB or more). The limited memory specifications of sensing devices can pose challenges when performing resource-intensive tasks or processing large datasets. Developers must carefully optimize algorithms and applications to work efficiently within the memory constraints of these devices. On the other hand, PCs have ample memory and storage capabilities, making them better suited for more computationally demanding tasks and data-intensive applications. It is important to note that the hardware properties can vary significantly across different devices, models, and manufacturers. The specifications and capabilities of the devices are constantly evolving, with newer generations offering more powerful CPUs, GPUs, increased memory capacities, and additional features along with various devices, including IoT devices, wearables, smartphones, robots, drones, and many more. For instance, IoT devices typically possess single or dual-core processors with limited memory and storage. At the same time, wearables like smart watches are equipped with low-power CPUs, moderate memory, and specialized health-related sensors. On the other hand, smartphones feature multi-core processors, substantial memory, high-resolution cameras, and expandable storage options. Robots may vary in CPU configurations and memory capacity, relying on various sensors for perception and navigation tasks. Drones often have specialized flight controllers, moderate memory, and microSD card storage. Considering the hardware properties of these devices is essential when developing applications and algorithms to ensure optimal performance and compatibility with the target hardware platform. In this review, our main emphasis lies on sensors commonly found on edge devices, such as motion sensors (accelerometer, gyroscope, and magnetic field sensors), location sensors (GPS and wireless interfaces), pressure sensors, thermometers, electrodermographs (EDA), electromyography (EMG), electroencephalographs (EEG), electrocardiographs (ECG), oximeters, and proximity sensors (e.g., Bluetooth). These sensors are frequently employed in various sensing applications, often in combination, resulting in the generation of multi-modal sensor data. In the medical field, sensors such as electrocardiogram (ECG) sensors are used to measure and record the heart’s electrical activity, aiding in diagnosing heart conditions. Pulse oximeters, another medical sensor, monitor oxygen saturation levels in the blood. In the environmental domain, water quality sensors are utilized to assess parameters like pH levels and dissolved oxygen in bodies of water. Gas sensors are employed for indoor air quality monitoring and detecting the presence of harmful gases. Automotive applications involve light sensors that adjust headlight and dashboard display brightness based on ambient light levels and rain sensors that automatically activate windshield wipers in response to rain. In industrial settings, flow sensors measure the flow rate of liquids or gases in pipelines, facilitating flow control and monitoring. Vibration sensors are utilized to detect vibrations in machinery, identify potential faults, and prevent breakdowns. Finally, in the aerospace sector, gyroscopes are used to measure angular motion, ensuring stability and navigation accuracy. These examples demonstrate the diverse range of sensors utilized in various domains to detect and measure environmental changes, enabling precise monitoring, control, and decision-making processes. This paper is available on Arxiv under a CC 4.0 license. This paper is available on Arxiv under a CC 4.0 license. available on Arxiv

Sensors and Edge-Sensing Devices: The Ways They Can Improve Our Lives

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

5 Ways to Measure Privacy Risks in Text Data

A Beginners Guide to Federated Learning

A Quick Introduction to Federated Learning of Cohorts [FloC]

A Zero-Knowledge Anomaly Detection Approach for Robust Federated Learning

Building Sustainable AI/ML Solutions in the Cloud with Federated Learning

5 Ways to Measure Privacy Risks in Text Data

A Beginners Guide to Federated Learning

A Quick Introduction to Federated Learning of Cohorts [FloC]

A Zero-Knowledge Anomaly Detection Approach for Robust Federated Learning

Building Sustainable AI/ML Solutions in the Cloud with Federated Learning

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps