Sudip Misra

Federated Learning (FL) is increasingly important in Internet of Things (IoT) for healthcare, as it facilitates decentralized training of machine learning models while ensuring privacy of sensitive data. However, FL faces difficulties with pathological non-Independent and Identically Distributed (non-IID) setting, where differences in data from various sources can reduce the effectiveness of the models. This work proposes SeAFed , a client ranking-based Federated Learning (FL) system designed for pathological non-IID environment . Current FL systems are often challenged with non-IID data, leading to inefficient and biased model training. SeAFed addresses these limitations by employing a server that coordinates with the clients using a ranking algorithm that prioritizes clients based on disease severity. SeAFed deploys client ranking using an Analytic Hierarchy Process (AHP) on the server by utilizing several criteria such as disease severity level, local accuracy, training time, and data size. This ranking demonstrates the contribution that each client provides to global accuracy. Using this approach, any number of clients can be selected after the ranking for training. We evaluate the system's performance in terms of accuracy, process time, CPU usage, and memory consumption. Our approach achieves 22.5% less process time, 29.8% less CPU usage and 19.5% less memory consumption. We observe 70% test accuracy on the non-IID data for all the clients without applying SeAFed as well as for 3 clients after applying SeAFed . We find that SeAFed achieves comparable accuracy while blackucing the resource consumption. Therefore, our approach improves the efficiency of FL in healthcare by effectively utilizing critical data to improve model performance.

This work proposes a marketplace model, EMsaaS , for sharing microservices on the IoT network on edge. Containerized microservices are well-suited for real-time IoT applications on edge because of their lightweight nature, independent lifecycle, and scalability. However, deploying microservices on demand is often failure-prone because of edge bandwidth limitations and resource constraints. Additionally, centralized image registries hosting containerized application images are often far away from the edge site and incur latency for the download of the images. They are also prone to a single point of failure. This work proposes a decentralized mechanism for sharing microservices in the edge network using a peer-to-peer approach. We introduce a component Edge Microservice Tracker and deploy it on the participating edge nodes. This component communicates with the other peer nodes and discovers microservices in the network. When requested, it fetches the requested microservice from the host node and deploys the same on the requesting edge node. We implement the proposed system in an IoT network of edge devices and evaluate the model using several parameters such as the deployment time, CPU and memory consumption, internet upload/download rate, and metadata synchronization time for sharing microservices on the edge. We observe that the system performs better than the cloud or centralized registry based approach with respect to the microservice deployment time with almost a 30% reduction in the deployment time even without any bandwidth limit for internet download from the edge node. It further surpasses the cloud registry-based microservice deployment procedure, with edge nodes having bandwidth limits. Moreover, EMsaaS proposes the concept of microservice sharing in an edge network, which enhances collaboration between edge service providers and fosters faster innovations.

Data fragmentation across IoT healthcare devices makes identifying and tracing causal health events challenging. In this work, we propose CasFly, a decentralized, on-the-fly causal chain tracing protocol for healthcare event sequences. We design a specialized data structure, temporal probabilistic health graph (TPHG), to dynamically trace event dependencies across heterogeneous healthcare devices. When a device detects an anomalous parameter at the edge, it queries its local TPHG to identify the most probable temporal precursors. We propose incremental chain formation at each of the devices using the lag-aware Viterbi expansion (LaVE) algorithm. The algorithm transmits data only when new causal paths are identified, thus reducing the latency of the dependency path construction. CasFly operates device-to-device by expanding the chain until no further expansion is possible. Finally, the chain is returned to the initiating device for visualization and analysis. We observe that on emulated IoT hardware, TPHGs load in $\leq$ 5 s using ~170 MB RAM. TPHGs reduce storage by ~97.6% versus conditional probability tables (CPTs). As compared to PCMCI+, LaVE improves macro AUPRC by $\approx 18$ % and reduces log-loss by $\approx 8$ %, with similar AUROC. These results indicate that on-the-fly chain construction supports real-time monitoring and clinician interpretation of probable cross-device event sequences.

Charging of Electric Vehicles (EVs) from stationary roadside charging points (CPs) that are connected to a smart grid constitute as unique transaction events. These events can thus be verified by and stored over a distributed blockchain over the grid, where the smart CPs themselves act as miners. However, mining being an inherently resource intensive task, the grid is expected to remain fairly loaded due to a high EV charging density. Hence, towards smooth EV adoption, in this work, we propose to reduce relevant mining overheads for such a blockchain over smart EV-utility grid using geographic segmentation of the miner set. Depending on the physical locations of the CPs and their connectivity graph, we divide an initially unsegmented mining cluster into several smaller clusters geographically, where each cluster verifies only their local transactions. This significantly reduces important mining parameters per-transaction and per-block, while achieving parallelism of block generation across segments. We argue that for blockchain transactions like EV charging that are coupled to their geographic source locations, it is sufficient to have smaller independent, but parallel mining segments. Through a detailed analysis, we also prove that reducing the mining segment size does not necessarily compromise on the blockchain security. We corroborate our claims through extensive experimental results, where the proposed solution achieves an average improvement of 48% with respect to CPU cycles expended and mining energy consumed per block and 45% with respect to mining time per block with half-sized segments as compared to an unsegmented miner set. Our experimental results further prove that the proposed segmentation outperforms other geographic clustering such as K-Means clustering for blockchain applications. Additionally, owing to parallel block generation, we show that by segmentation, the mining energy required per block is reduced from 19 J to about 0.02 J over the same grid, providing a significant energy efficiency.

Gas-sensing systems and their networks play a critical role in ensuring safety and environmental compliance across industrial and residential environments. Their reliability, however, depends on the rapid and accurate detection of gas leaks and abnormal gas levels. This article addresses the challenge of limited selectivity in gas sensor nodes by developing next generation, Internet of Things (IoT)-enabled air quality monitoring (AQM) nodes. These nodes leverage intelligent machine learning models to form an edge-assisted gas-sensing network (Ea-GNet). Using low-cost metal–oxide–semiconductor (MOS) chemical sensors, the proposed system enables selective detection and accurate identification of multiple gases from complex mixtures, followed by precise concentration estimation. Experimental results demonstrate a classification accuracy of approximately 98% on randomly mixed samples and an average ${R}^{{2}}$ -score of 0.84 for volatile organic compounds (VOCs), such as acetone, ethanol, and methanol. In addition, the integration of edge computing significantly improves the processing efficiency, supporting minimal delays and enabling near-real-time monitoring. With the adopted simulation settings, the edge computing improved the throughput of the Ea-GNeT system by 634% compared with the cloud-based processing with fixed quality of service (QoS) settings. The latency of Ea-GNet is remarkably reduced from the order of 18 s–52 ms when the number of sensor nodes reaches 200. The findings highlight the viability of implementing intelligent machine learning models at the edge to improve the responsiveness and reliability of gas-sensing networks.

Advancements in Vehicular Ad-Hoc Networks (VANETs) are crucial for the next generation of intelligent transportation systems. Addressing the complexities of routing and clustering in these networks, our research introduces the Multi-constraint Routing Mechanism using Hybridization (MRMH). This approach innovatively combines the strengths of Grey Wolf Optimization (GWO) and Sequential Quadratic Programming (SQP). While GWO is adept at global search, its tendency for premature convergence is effectively countered by SQP's excellence in nonlinear constraint management and local optimization. MRMH further benefits from a novel weighted distance approach and a nonlinear decay formulation, enhancing the balance between exploration and exploitation phases in VANET optimization. Our fitness function, designed around essential VANET metrics like inter-cluster distance and vehicle orientation, ensures the applicability of MRMH in real-world scenarios. Empirical results from extensive simulations exhibit MRMH's superiority, showing significant improvements in key performance indicators such as transmission delay (16% reduction), packet delivery ratio (12% increase), throughput (18% increase), cluster stability (23% improvement), and convergence rate (11% increase). These findings are backed by rigorous statistical analyses, confirming MRMH's potential in VANETs.

The distributed Internet of Things (IoT) systems facilitate real-time services through the composition of loosely coupled microservices. The composed IoT service is the output of multiple microservices, executed at computationally capable edge or fog nodes, which we consider as the facility nodes (FNs). However, the service composition process in IoT microservice architectures is abstracted from the users that gives freedom to the FNs to act maliciously and provide low-quality IoT services. Further, the service composition needs to be transparent so that the FNs involved in a service cannot repudiate their involvement at a later time. In this article, we propose a novel, lightweight, trustworthy, and verifiable service composition framework for IoT-based systems that adopt microservice architecture. First, we propose a dynamic programming approach to select trustworthy FNs for each user request, while considering the trust scores of the FNs and the delay requirements of the users. Next, we propose a transparent service composition framework that uses lightweight cryptography functions to generate the proof-of-involvement for the FNs in each service. With the help of a trust controller, we verify the proofs generated by the FNs and update the trust scores of the FNs. Considering the user traces from Berlin city in the simulation of urban mobility tool, we show the efficacy of the proposed framework in maximizing user trust and detecting malicious FNs involved in user services. Further, we show that the delay and communication overhead of the proposed framework are very low compared to the state-of-the-art methods.

This work addresses the problem of ensuring service availability, trust, and profitability in sensor-cloud architecture designed to Sensors-as-a-Service (Se-aaS) using IoT generated data. Due to the requirement of geographically distributed wireless sensor networks for Se-aaS, it is not always possible for a single Sensor-cloud Service Provider (SCSP) to meet the end-users requirements. To address this problem, we propose a federated sensor-cloud architecture involving multiple SCSPs for provisioning high-quality Se-aaS. Moreover, for ensuring trust in such a distributed architecture, we propose the use of consortium blockchain to keep track of the activities of each SCSP and to automate several functionalities through Smart Contracts. Additionally, to ensure profitability and end-user satisfaction, we propose a composite scheme, named BRAIN, comprising of two parts. First, we define miner's score to select an optimal subset of SCSPs as miners periodically. Second, we propose a modified multiple-leaders-multiple-followers Stackelberg game-theoretic approach to decide the association of an optimal subset of SCSPs to each service. Thereafter, we evaluate the performance of BRAIN by comparing with three existing benchmark schemes through simulations. Simulation results depict that BRAIN outperforms existing schemes in terms of profits and resource consumption of SCSPs, and price charged from end-users.

The heterogeneity in IoT-based health monitoring devices and communication protocols leads to diverse system designs and configurations resulting in uncertain outcomes. Most present-day eHealth approaches focus on optimizing the performance of already designed systems (hardware, network, and software) rather than optimizing the system design itself. A reliable eHealth monitoring system must use appropriate protocols and hardware to avoid data loss and delays. Identifying the magnitude of performance changes when integrating new devices into an already functional eHealth system is also critical. This work addresses these issues by adopting a queuing theory-based analytical model to characterize an eHealth wearable system. This approach acts as a guiding framework for system and network capacity estimation by simulating the dynamics of an eHealth wearable system with selected communication protocols before deployment. The simulation tool identifies the system's limits in terms of delays, the number of simultaneously supported wearables, and packet loss. We identify and examine some well-known wireless connectivity protocols as possible candidates for eHealth wearable systems. The output of the analytical model is compared vis-à-vis data from a real-life proof-of-concept eHealth wearable system. The proposed approach estimates the numbers of served packets and blocked packets with approximate errors of 0.02% and 0.2%, respectively.