Graph Neural Networks in IoT: A Survey

Abstract

The Internet of Things (IoT) boom has revolutionized almost every corner of people’s daily lives: healthcare, environment, transportation, manufacturing, supply chain, and so on. With the recent development of sensor and communication technology, IoT artifacts including smart wearables, cameras, smartwatches, and autonomous systems can accurately measure and perceive their surrounding environment. Continuous sensing generates massive amounts of data and presents challenges for machine learning. Deep learning models (e.g., convolution neural networks and recurrent neural networks) have been extensively employed in solving IoT tasks by learning patterns from multi-modal sensory data. Graph neural networks (GNNs), an emerging and fast-growing family of neural network models, can capture complex interactions within sensor topology and have been demonstrated to achieve state-of-the-art results in numerous IoT learning tasks. In this survey, we present a comprehensive review of recent advances in the application of GNNs to the IoT field, including a deep dive analysis of GNN design in various IoT sensing environments, an overarching list of public data and source codes from the collected publications, and future research directions. To keep track of newly published works, we collect representative papers and their open-source implementations and create a Github repository at GNN4IoT.

FAQ: Feature Aggregated Queries for Transformer-based Video Object Detectors

“…In other work like [69,70,77], NAS is applied to the existing image object detectors to improve the performance. Besides NAS, recent works have begun using dynamic models [13,21,51,62,81,84,89] to improve efficiency dynamic convolutions [7,16,17,72], dynamic heads [12], and dynamic proposals [10], which are introduced to the existing object detectors to balance the performance and inference speed. For video object detection, DFA [8] proposes a model which can dynamically select the frames used to aggregate the features according to the input frames.…”

Artificial intelligence of things (AIoT) data acquisition based on graph neural networks: A systematical review

“…The existing surveys on progress in AI research for IoT are presented from four perspectives: perceiving, learning, reasoning, and behaving in Reference74 A comprehensive review of recent advances in the application of GNNs to the IoT field is given in Reference15 in which three typical IoT sensing paradigms and their combinations are reviewed: autonomous thing, human‐centric, and environment embedded systems. GNN for anomaly detection in three typical Industrial IoT (IIoT) fields are examined in Reference75 including smart transportation, smart energy and smart factory.…”

“…It should be explicitly pointed out that the existing GNN works in IoT sensing applications have been surveyed by Dong et al, 15 including three typical of IoT sensing scenarios: human sensing, autonomous things, and environmental sensing. Significantly different from the work above, instead of presenting the various GNN schemes in vertical IoT sensing applications, our work focuses on the systematically reviewing the horizontal infrastructure that all AIoT applications should possess, that is, data sensing and acquisition based on GNN and some emerging AI factors including adversarial training, self‐supervision learning, and graph reinforcement learning, etc.…”

Semi-supervised Graph Instance Transformer for Mental Health Inference

“…We hypothesize that people who are in early stage of influenza will subconsciously become less active, and avoid unnecessary traveling and person-to-person interactions. We leverage multiple embedded sensors to capture human behavior information: GPS sensors are used to capture human mobility behaviors; Bluetooth encounters are used as proximity to measure social behaviors; accelerometer and gyroscope can monitor physical activeness; web-virtual behaviors are tracked by recording app usages; and ambient contexts can be inferred by WiFi signals [7].…”

“…Many privacy-preserving approaches and algorithms have been proposed according to different application scenarios. For instance, k-anonymity algorithm and its variants have been proposed to hide identities in the source of data items [7,8]. Differential privacy aims to reduce exposure of authentic information by adding perturbations that follow certain noise distribution while maintain data utility.…”