This paper presents a computationally efficient semi-supervised learning (SSL) framework designed for real-time, on-device learning in resource-constrained IoT-based health monitoring systems. The proposed methods focus on minimizing redundant computation and memory usage, which are directly linked to energy consumption in embedded platforms. First, Mini-Batch K-Means clustering is employed as an alternative to full-batch clustering, reducing per-iteration computational time and memory footprint, particularly for low to moderate-dimensional datasets. To further enhance efficiency, two unsupervised convergence detection mechanisms, β-stop and ζ-stop are introduced to autonomously halt clustering once model stability is achieved, preventing unnecessary retraining and reducing energy overhead. β-stop monitors the stabilization of clustering iterations per learning cycle, while ζ-stop tracks the rate of cluster growth as a convergence indicator. Experimental evaluations on three representative IoT health monitoring datasets:Smart Hydration Tracking (SHT), Human Activity Detection (HAD), and Infant Activity Detection (IAD)demonstrate that the proposed strategies reduce computational time by up to 45% and CPU memory consumption by up to 30% without compromising classification accuracy. The results confirm the framework’s scalability, energy efficiency, and suitability for reliable, real-time semi-supervised learning on embedded IoT devices.

This paper presents a deep learning framework for energy-and bandwidth-aware time-series regression for IoT and wireless sensor networks. Leveraging cross-modality structural data dependencies, the approach uses a multivariate regression approach for predicting many different sensor modality time-series from a fewer number of sensor modality data. This reduces the communication bandwidth usage and energy consumption for resource-constrained sensor nodes. In addition, the approach can be leveraged for reducing number of sensors in a system, thus reducing the hardware and associated costs in IoT-based systems. This is done while ensuring successful reconstructions of all the original time-series for the necessary sensing modalities. One key attribute of the proposed framework is that the processing load is placed entirely at the receiver side, thus making it suitable for resource-constrained sensors and IoT based systems. The developed methodology is validated using data collected by a greenhouse sensor IoT system deployed over a period of three years. The results demonstrate the ability of the proposed approach for successful prediction of many sensor modality times series from only one time-series, representing the energy harvesting super-capacitor voltage of the sensor IoT system. The trade-offs between performance, bandwidth usage, number of sensing modalities, scalability, and computation cost are analyzed for providing design and implementation guidelines.

This paper introduces an Asymmetric Autoencoder (AAE)-driven data compression framework for efficient management of energy, bandwidth, and transmitter complexity in a Wireless Sensor Network (WSN). WSNs are often limited by their ability to process and transmit high-dimensional data due to various constraints, including available energy, processing cycles, and transmission capacity. Achieving an application-specific downstream task executed in a remote receiver under the influence of such sensor node constraints is the focus of the proposed methodology. It places a data compression encoder and decoder at the transmitter and the receiver, respectively. The architecture of the AAE's encoder and decoder can be asymmetric, and the degree of asymmetry can be adjusted based on the computation and processing abilities of the transmitting node, the available bandwidth, and the performance requirements of a specific downstream task. In the proposed framework, the encoder and decoder are jointly trained, enabling the system to extract downstream task-related information from one or more time series inputs. Verified for Human Activity Recognition (HAR), the framework demonstrates effective feature compression while maintaining efficient task performance.

This article introduces an Unmanned Aerial Vehicle - enabled content management architecture that is suitable for critical content access in communities of users that are communication-isolated during diverse types of disaster scenarios. The proposed architecture leverages a hybrid network of stationary anchor UAVs and mobile Micro-UAVs for ubiquitous content dissemination. The anchor UAVs are equipped with both vertical and lateral communication links, and they serve local users, while the mobile micro-ferrying UAVs extend coverage across communities with increased mobility. The focus is on developing a content dissemination system that dynamically learns optimal caching policies to maximize content availability. The core innovation is an adaptive content dissemination framework based on distributed Federated Multi-Armed Bandit learning. The goal is to optimize UAV content caching decisions based on geo-temporal content popularity and user demand variations. A Selective Caching Algorithm is also introduced to reduce redundant content replication by incorporating inter-UAV information sharing. This method strategically preserves the uniqueness in user preferences while amalgamating the intelligence across a distributed learning system. This approach improves the learning algorithm's ability to adapt to diverse user preferences. Functional verification and performance evaluation confirm the proposed architecture's utility across different network sizes, UAV swarms, and content popularity patterns.

This paper presents a Micro-Unmanned Aerial Vehicle (UAV)-enhanced content management system for disaster scenarios where communication infrastructure is generally compromised. Utilizing a hybrid network of stationary and mobile Micro-UAVs, this system aims to provide crucial content access to isolated communities. In the developed architecture, stationary anchor UAVs, equipped with vertical and lateral links, serve users in individual disaster-affected communities. and mobile microferrying UAVs, with enhanced mobility, extend coverage across multiple such communities. The primary goal is to devise a content dissemination system that dynamically learns caching policies to maximize content accessibility to users left without communication infrastructure. The core contribution is an adaptive content dissemination framework that employs a decentralized Top-k Multi-Armed Bandit learning approach for efficient UAV caching decisions. This approach accounts for geo-temporal variations in content popularity and diverse user demands. Additionally, a Selective Caching Algorithm is proposed to minimize redundant content copies by leveraging inter-UAV information sharing. Through functional verification and performance evaluation, the proposed framework demonstrates improved system performance and adaptability across varying network sizes, micro-UAV swarms, and content popularity distributions.

This paper proposes a learning-driven approach for medium access slot allocation in the presence of malicious nodes. Learning policies are developed with the goal of defending against several forms of quasi-random slot-scheduling attack models used by the malicious nodes. The primary learning objective for the non-malicious nodes is to minimize the degradation in network performance caused by the malicious nodes. This is accomplished while minimizing the bandwidth share of the malicious nodes. These objectives are achieved using a Multi-Armed Bandit (MAB) learning architecture that allows the nodes to learn transmission schedule on-the-fly, and without the need for any central arbitrator. Two different scheduling policies are introduced: robust and reactive policies. Following the design, a detailed characterization of these policies and their use in different application-specific scenarios are presented. An analytical model of the system is developed to find the benchmark throughput for different malicious attack models. It is demonstrated that the proposed framework allows network nodes to learn close-to -benchmark slot scheduling, while thwarting attacks from the malicious nodes. The proposed architecture is validated for various mesh networks and traffic conditions in the presence of different attack models enacted by the malicious nodes.

This paper proposes an unmanned aerial vehicle (UAV) aided content management system in communication-challenged disaster scenarios. Without cellular infrastructure in such scenarios, community of stranded users can be provided access to situation-critical contents using a hybrid network of static and traveling UAVs. A set of relatively static anchor UAVs can download content from central servers and provide content access to its local users. A set of ferrying UAVs with wider mobility can provision content to users by shuffling them across different anchor UAVs while visiting different communities of users. The objective is to design a content dissemination system that on-the-fly learns content caching policies for maximizing content availability to the stranded users. This paper proposes a decentralized Top-k Multi-Armed Bandit Learning model for UAV-caching decision-making that takes geo-temporal differences in content popularity and heterogeneity in content demands into consideration. The proposed paradigm is able to combine the expected reward maximization attribute and a proposed multidimensional reward structure of Top-k Multi-Armed Bandit, for caching decision at the UAVs. This study is done for different user-specified tolerable access delay, heterogeneous popularity distributions, and inter-community geographical characteristics. Functional verification and performance evaluation of the proposed caching framework is done for a wide range of network size, UAV distribution, and content popularity.

This paper presents a lightweight on-device liquid consumption tracking system based on a semi-supervised learning paradigm. The online learning framework caters to scenarios where a hydration tracking bottle/device has no prior knowledge of a user's consumption gesture patterns. The proposed iterative semi-supervised learning (ISSL) framework uses sparsely labelled user gesture events acquired from the IMU sensors installed on a bottle, such that it can learn to differentiate between sip and non-sip gestures by specific individuals. Two different strategies, namely, population-based, and distance-based, are employed to achieve the desired clustering performance. A comparative study between these strategies has been presented in terms of clustering accuracies for classifying sip and non-sip gestures. The proposed architecture is shown to be lightweight in terms of computation complexity and memory usage of the bottle-embedded hardware. The trade-off between classification accuracy and computation complexity is analyzed for different algorithmic hyper-parameters and it is shown how to manage this trade-off. Extensive experimentation and simulation study has been conducted for multiple users' drinking patterns to validate the proposed learning paradigm.

This article proposes an unmanned aerial vehicle (UAV) aided content provisioning system in communication-challenged disaster scenarios. In such scenarios, without the availability of static base stations and their wireline backhauls, community of stranded users can access contents from a network of static and traveling UAVs. A set of relatively static anchor UAVs with vertical as well as lateral links can provide content access to its local users. A set of ferrying UAVs with only lateral links, but with wider mobility, can provision content to users while visiting different communities of users. The objective is to design a content dissemination system that handles user demand heterogeneity while maximizing content availability to the requesting users. This work proposes a popularity-based caching policy which tackles heterogeneity in content popularity across a disaster region. A novel value-based caching policy is developed which considers the popularity and the tolerable access delay of the content requests from the users to make caching decisions. The paper develops a novel approach called Joint Deployment of Ferrying UAVs (JDFU) Algorithm to exploit the collective storage of ferrying UAVs which boosts content distribution for users. Through analytical modeling and simulation experiments it is demonstrated that content availability can be maximized by choosing an optimal cache storage segmentation factor, JDFU configuration and hover time of ferrying UAVs. This analysis is done for different user-specified tolerable access delay, heterogeneous popularity distributions and intercommunity geographical characteristics. The paper does functional verification and performance evaluation of the proposed caching framework under a wide range of network size, UAV distribution, content popularity, and ferrying UAV trajectories.

This article presents an unmanned aerial vehicle (UAV) based caching framework for content provisioning in disaster scenarios. In a disaster scenario without the availability of static base stations and their wireline backhauls, community of stranded users can access contents from a network of static and traveling UAVs. A set of relatively static anchor UAVs with vertical as well as lateral links provide content access to its local users. A set of ferrying UAVs with only lateral links, but with wider mobility, can also provision content to users while visiting different communities of users. The algorithmic objective is to intelligently cache contents in the storage-constrained UAVs in order to maximize content availability for the users affected by such disasters. The paper develops a novel approach of content duplication within the anchor UAVs along with a mechanism to distribute non-duplicated contents across the ferrying UAVs. Through analytical modeling and simulation experiments it is demonstrated that content availability in such an arrangement can be maximized by choosing an optimal level of duplication for content with specific popularity distributions. The paper does functional verification and performance evaluation of the proposed caching framework under a wide range of network size, UAV distribution, content popularity, and ferrying UAV trajectories.

This paper presents a distributed framework for Medium Access Control (MAC) slot allocation in time-asynchronous wireless networks using Multi-Armed Bandits (MAB) based learning. MAC slot allocation is formulated as an MAB problem where the nodes act as independent learning agents and learn transmission policies that ensure collision free transmissions. A novel concept of Hysteretic MAB has been introduced to speed up learning convergence. In order to reduce the bandwidth overhead while maintaining a desired MAB learning speed, a novel slot defragmentation mechanism is introduced. Not relying on network time synchronization makes the proposed mechanism feasible for low-complexity and low-cost transceivers for wireless sensor and loT networks. The proposed mechanism is tested and evaluated on both fully connected and arbitrary mesh network topologies and is shown to be scalable with network size and topological degree. It is also shown that in partially connected topologies, the mechanism learns spatial channel reuse, thus leading to better spectral usage efficiency.