Intelligent Transportation Systems (ITS) demand reliable vehicle identification and overload detection to enable proactive traffic management, enhancing infrastructure durability and road safety. Prior studies have employed unimodal approaches that primarily rely on visual imagery/video or sensor data, which often struggle in real-world environments due to challenges like occlusions, variable lighting, and sensor noise. To address such challenges, we propose VehicleSense, an adaptive, real-time, and multi-modal learning framework that fuses visual and inertial data for robust vehicle identification and overload detection. The framework integrates image-based (ImageSense) and sensor time-series-based (SensorSense) modalities using a trainable neural meta-classifier. We further introduce a late-fusion strategy to integrate predictions from both modalities, enabling dynamic weighting and conflict resolution to improve robustness. This fusion approach enhances classification reliability, ensuring adaptability across diverse environmental conditions and operational scenarios. Comprehensive evaluations on the collected real-world multi-modal dataset demonstrate that VehicleSense significantly outperforms unimodal baselines, achieving substantial gains. In addition, the framework is optimized for edge deployment, enabling scalable, real-time vehicle intelligence in resource-limited settings.

Adversarial robustness in Unsupervised Domain Adaptation (UDA) remains a significant challenge due to noisy pseudo labels and inherent distributional shifts between the clean source and adversarially perturbed target domains. Existing approaches often fail to achieve an optimal trade-off between robustness and accuracy, as pseudo-labels generated by domain-adapted models tend to introduce classification errors under adversarial attacks. In this work, we propose SFT+RL, a two-stage robust UDA framework that integrates Supervised Fine Tuning (SFT) and Reinforcement-inspired Learning (RL) on top of CLIP’s pre-trained visual encoder. In the SFT stage, we adversarially fine-tune a linear classifier using PGD-based perturbations over the labelled source domain while partially unfreezing CLIP’s projection layer. It allows adaptation to adversarial noise while preserving CLIP’s rich semantic priors. We introduce a confidence-guided pseudo-labeling strategy in the RL stage to annotate unlabeled target samples progressively. Pseudo labels are filtered using a decaying confidence threshold to balance quality and coverage, and the model is trained on a composite dataset formed by combining clean source samples with high-confidence target samples. Adversarial training is applied to mixed batches of clean and adversarial examples to enhance cross-domain robustness. Comprehensive evaluations on three benchmark datasets OfficeHome [18], PACS [7], and VisDA [13] demonstrate the effectiveness of our approach. Notably, SFT+RL achieves average improvements of 10.2% in clean accuracy and 15.8% in adversarial robustness across all three datasets, outperforming existing state-of-the-art methods.

LoRaWAN has emerged as a leading LPWAN technology for emergency response systems due to its energy efficiency and wide coverage. However, its limited bandwidth poses significant challenges for transmitting time-sensitive images, which are crucial for first responders. A major constraint in such scenarios is that the sender - typically a resource-constrained edge device - cannot perform complex Machine Learning (ML) operations for image compression or enhancement prior to transmission. Conversely, the receiver, often a more capable gateway, has sufficient computational resources to handle ML-based image reconstruction. To address this asymmetry, we propose ZML-ChunkLoRa, an adaptive image transmission framework that segments images into optimally sized chunks based on real-time network conditions. Reduce sender-side processing while enabling high-quality ML-based reconstruction at the receiver. By offloading computation to the gateway, ZML-ChunkLoRa improves transmission efficiency without violating the strict energy and bandwidth restrictions of LoRaWAN. Experimental results demonstrate that our approach significantly improves image transfer speed and visual quality in time-critical, resource-limited first-responder environments.

The popularity of wearable smart devices has increased due to their seamless monitoring of vital signs during daily activities. Federated learning leverages these devices along with participants’ smartphones to fine-tune pre-trained models. Moreover, calibrating the differences between wearables and smartphones in terms of sampling rates, orientations, activity correlation, battery power, and other factors is challenging. Thus, the paper introduces a participant and wearable selection cross-device federated learning approach. It leverages criteria such as the activity wearable(s) relationship, data quality, battery life, sampling rate, and so on to perform the wearable selection. The server evaluates and estimates the utility of each participant and selects those with higher utility in each communication round. We then figure out the optimal weighted contribution of each participant to perform robust aggregation. We also use knowledge distillation techniques to develop a high-performing and lightweight wearable model. Finally, we conduct simulation and real-world experiments on existing datasets and compare our approach with state-of-the-art. The result shows an improvement of$3\!\!-\!\!4\%$in accuracy via fine-tuning from selected wearable data.

Federated learning enables multiple participants to collaboratively train machine learning models without the need to share their private and limited data, thereby preserving privacy. When datasets used in federated learning contain noisy labels, it can lead to degraded performance and an increased number of communication rounds needed to achieve convergence. This, in turn, requires more time and energy to train the model. This paper proposes a federated learning approach to suppress the unequal distribution of the noisy labels in the dataset of each participant. The approach first estimates the noise ratio of the dataset for each participant and normalizes it using the server dataset. Next, the approach considers the influence of each participant and calculates the optimal weighted contributions for each one. The approach also considers bias in the server dataset and minimizes its impact on the participants. Further, the paper provides an expression to estimate the number of communication rounds required for convergence. Results demonstrate the superiority of the proposed approach over baselines in terms of communication rounds and performance.

Internet of Things (IoT) employs pervasive integration of sensors-equipped entities, such as devices, vehicles, and home appliances. These entities collect and exchange data using application-specific software and require seamless network connectivity. The seamless communication between IoT devices enables remote monitoring and control via the Internet. However, the reliability of communication comes at the cost of device and network resources, adding strain to the resource-constrained IoT environment. In this paper, we propose a novel and efficient approach to attain user-specified reliability in IoT-enabled sensor networks. Apart from being statically confined to a single communication protocol, the approach dynamically selects between the transmission control-based Message Queuing Telemetry Transport (MQTT) and user datagram-based MQTT protocols. This ensures optimal resource utilisation of the IoT devices and communication networks. The approach considers the reliability fraction as the input from the users and accesses the IoT device and network resources, including processing power, memory, and network speed to choose the optimal protocol for transmitting the data in the network. Furthermore, we employ the long short-term memory on the time series data of IoT device resources to decide the most suitable protocol for data transmission. The decision serves as feedback to users for reevaluating their reliability requirements. Through experimental evaluation, we validate the effectiveness of our approach in terms of resource efficiency and communication time with the given reliability.

Federated learning is a distributed training framework that ensures data privacy and reduces communication overhead to train a shared model among multiple participants. Noise in the datasets and communication channels diminishes performance and increases communication rounds for convergence. This paper proposes a federated learning approach to rectify noise in local datasets and communication channels for reducing communication rounds. We employ two filters: one to rectify noise in the dataset and the other for the communication channel. We also consider two types of participants: one uses a filter on the dataset, and the other uses both the dataset and the communication channel. We first derive the expression to estimate the communication rounds for convergence. Next, we determine the number of participants under each type. Finally, we perform the experiments to verify the effectiveness of the proposed work on existing datasets and compare them with state-of-the-art techniques.

Automated feature extraction capability and significant performance of Deep Neural Networks (DNN) make them suitable for Internet of Things (IoT) applications. However, deploying DNN on edge devices becomes prohibitive due to the colossal computation, energy, and storage requirements. This paper presents a novel approach, EarlyLight, for designing and training lightweight DNN using large-size DNN. The approach considers the available storage, processing speed, and maximum allowable processing time to execute the task on edge devices. We present a knowledge distillation based training procedure to train the lightweight DNN to achieve adequate accuracy. During the training of lightweight DNN, we introduce a novel early halting technique, which preserves network resources; thus, speedups the training procedure. Finally, we present the empirically and real-world evaluations to verify the effectiveness of the proposed approach under different constraints using various edge devices.

Federated learning is an emerging paradigm that provides privacy-preserving collaboration among multiple participants for model training without sharing private data. The participants with heterogeneous devices and networking resources decelerate the training and aggregation. The dataset of the participant also possesses a high level of variability, which means the characteristics of the dataset change over time. Moreover, it is a prerequisite to preserve the personalized characteristics of the local dataset on each participant device to achieve better performance. This article proposes a model personalization-based federated learning approach in the presence of variability in the local datasets. The approach involves participants with heterogeneous devices and networking resources. The central server initiates the approach and constructs a base model that executes on most participants. The approach simultaneously learns the personalized model and handles the variability in the datasets. We propose a knowledge distillation-based early-halting approach for devices where the base model does not fit directly. The early halting speeds up the training of the model. We also propose an aperiodic global update approach that helps participants to share their updated parameters aperiodically with server. Finally, we perform a real-world study to evaluate the performance of the approach and compare with state-of-the-art techniques.

Federated Learning is a training framework that enables multiple participants to collaboratively train a shared model while preserving data privacy. The heterogeneity of devices and networking resources of the participants delay the training and aggregation. The paper introduces a novel approach to federated learning by incorporating resource-aware clustering. This method addresses the challenges posed by the diverse devices and networking resources among participants. Unlike static clustering approaches, this paper proposes a dynamic method to determine the optimal number of clusters using Dunn Indices. It enables adaptability to the varying heterogeneity levels among participants, ensuring a responsive and customized approach to clustering. Next, the paper goes beyond empirical observations by providing a mathematical derivation of the communication rounds for convergence within each cluster. Further, the participant assignment mechanism adds a layer of sophistication and ensures that devices and networking resources are allocated optimally. Afterwards, we incorporate a master-slave technique, particularly through knowledge distillation, which improves the performance of lightweight models within clusters. Finally, experiments are conducted to validate the approach and to compare it with state-of-the-art. The results demonstrated an accuracy improvement of over 3% compared to its closest competitor and a reduction in communication rounds of around 10%.

Patient activities’ monitoring is a promising application of the Internet of Medical Things (IoMT), revolutionizing clinical diagnosis. An IoMT uses sensory data collected from smart devices to train a model on the server. The trained model recognizes the patient activities on smart devices. However, training the model on the server has raised privacy concerns and security threats. The sensitive medical data transferred from the smart devices to the Cloud poses different cybersecurity challenges, such as distributed denial of service (DDoS), phishing, network penetration, and side-channel. This article proposes a secure patient monitoring system using federated learning (FL). The system performs training on local devices and sends only weight matrices to the server for aggregation; thus, it preserves data privacy and security compromises. The system intelligently divides the participants into clusters based on the available resources, trains suitable models on each cluster, and enhances the performance via knowledge distillation (KD). The model of high-performing clusters distills knowledge to the model of small-size clusters to improve their performance. The experimental results illustrate that the proposed system successfully work in the presence of unequal resources.

Augmented intelligence is an innovative amplification of artificial intelligence that allows human experts to take over the autonomous decision of machines. It also facilitates human-intelligence-based decisions on the network edge using low-cost and small-sized devices. Augmented intelligence and the Internet of Things collectively create augmented intelligence of things. It logically and effortlessly interrelates human intelligence to articulate smart decisions. Unmanned aerial vehicles find various applications ranging from search operations during disasters to intruders identification; thus, they are suitable for aerial networks, where connections between base stations and servers are extinct. This article presents an approach to enhance the lifetime of unmanned-aerial-vehicles-enabled aerial networks via augmented intelligence. It first considers the available battery power to transform a large-size deep neural network into a lightweight. We next present a knowledge-distillation-based approach, which reduces training time and enhances accuracy. Finally, we evaluate the approach on the existing dataset.

Restrictive public health measures such as isolation and quarantine have been used to reduce the pandemic virus's transmission. With no proper treatment, older adults have been specifically advised to stay home, given their vulnerability to COVID-19. This pandemic has created an increasing need for new and innovative assistive technologies capable of easing the lives of people with special needs. Smart home systems have become widely popular in providing such assistive services to isolated older adults. These systems can provide better services to assist older people if it anticipates what activities inhabitants will perform ahead of time. For example, a smart home can prompt inhabitants to initiate essential activities like taking medicine using activity prediction. This paper proposes a multi-task activity prediction system that jointly predicts labels, locations, and starting times of future activities. The observed sequence of previous activities characterizes future activities. We use body activity information from wearable sensors and motion information from passive environmental sensors to sense activities of daily living of older adults. The activity prediction system consists of recurrent neural networks to capture temporal dependencies. This work also carries out several experiments on collected and existing real datasets to evaluate the system's performance.

Intelligent Transportation System (ITS) helps to improve vehicle health, driver safety, and passenger comfort. Remotely sharing the information of ITS to train the machine and deep learning models hamper data privacy and generate security threats to the passenger, driver, and vehicle owners. Moreover, sharing the information requires huge networking resources such as high data rate, low latency, and low packet loss. Federated learning provides privacy-preserving model training on the vehicle without sharing the information. However, due to poor annotation mechanisms, federated learning may suffer from imperfect labels. This paper proposes a federated learning approach for ITS that can handle imperfect labels in the datasets of the participants. The approach also uses a Long-Range network to provide communication efficient connectivity. The approach initially estimates class-wise centroids of the datasets at the participants and server and then identifies participants with imperfect labels using similarity scores. Such participants demand the fraction of the correctly annotated dataset at the server to improve performance. We further derive the expression for the optimal fraction of the dataset requested by a participant. We finally verify the effectiveness of the proposed approach using the existing model and publicly available dataset.

Availability of various sensors in the smartphone makes it easier and convenient to collect the data of human locomotion activities. A recognition approach can utilize this sensory data for recognizing a locomotion mode of a user such as a bicycle, bike, car, etc. Such recognition of locomotion modes helps in the precise estimation of transportation expenditure, travel time, and appropriate journey planning. The accuracy of the recognition approaches heavily relies on the training dataset having correctly annotated labels. These labels are usually assigned using crowdsourcing or web-based queries for economic and fast annotation. However, the annotation generates abundant noisy labels in the dataset. This paper proposes a locomotion mode recognition approach capable of handling noisy labels in the training dataset. The approach builds an ensemble model by developing three different deep learning-based models, namely conventional, noise adaptive, and noise corrective, to handle different concentrations of noisy labels. The ensemble model not only improves the recognition performance but also helps in estimating the concentration of noisy labels. Experimental results demonstrate the effectiveness of the proposed approach on collected and existing datasets.

With the availability of low-cost and low-power sensors, it becomes easier to assess river water quality. The existing work on water quality assessment require a large amount of correctly annotated data for training. However, in the real-world scenario, obtaining such annotated data is costly and time consuming. In this work, we propose a sensor-based river water quality assessment system using the deep neural network (DNN). The system first presents a technique to estimate the water quality index (WQI) for labeling the given lab samples. WQI is a vital matrix used to transform large quantities of water data into a single unified number. Next, we present an automatic annotation technique that assigns labels to the sensory data instances using lab data. Finally, the labeled sensory data instances are used to build a DNN classifier that predicts water quality. This work also proposes a noise handling loss function to accommodate noisy labels. We evaluate the performance of the system on the river data set of major Indian rivers. We use four performance metrics during the experiment, including precision, recall, accuracy, and$F1$score. Additionally, the system achieves an accuracy of more than 90%, despite 20% noisy labels. The code is athttps://github.com/sourcecodecselab/river_water_monitoring.

Industrial IoT (IIoT) accomplishes digital manufacturing that incorporates various devices, simulators, and tools with multiple sensors. These sensors provide sufficient data to coordinate and monitor industrial systems. IIoT requires dedicated supporting devices for an application to execute a given task in maximum allowable response time. The requirement of dedicated devices increases system cost. Task containerization is a process of exploiting available resources of the host machines to meet out varying demands of different applications. It avoids the additional cost to buy dedicated IIoT devices while adding new applications. This article proposes an approach to securely process a given IIoT task within an allowable response time. We use the game theory approach to estimate the fractions of the task to be containerized on the machines. Next, the estimated fractions for each machine maximize the system utility. Finally, we illustrate the experimental results to validate the performance of the proposed approach.

With the availability of various sensors in the smartphone, identifying a locomotion mode becomes convenient and effortless in recent years. Information about locomotion mode helps to improve journey planning, travel time estimation, and traffic management. Though there exists a significant amount of work towards locomotion mode recognition, the performance of these work is not pertinent and heavily depends on the labeled training instances. As it is impractical to gather a prior information (labeled instances) about all types of locomotion modes, the recognition model should be able to identify a new or unseen locomotion mode without having any corresponding training instance. This paper proposes a sensors based deep learning model to identify a locomotion mode by using labeled training instances. The approach also incorporates a concept of Zero-Shot learning to identify an unseen locomotion mode. The model obtains an attribute matrix based on the fusion of three semantic matrices. It also constructs a feature matrix by extracting the deep learning and hand-crafted features from the training instances. Later, the model builds a classifier by learning a mapping between attribute and feature matrices. Finally, this work evaluates the performance of the approach on collected and existing datasets using accuracy and F1 score.

A Fog computing-based transportation system envisions to reduce energy consumption and communication delay. This paper presents a Fog computing-based scheme for assisting passengers, which involves task offloading and reallocation. We consider a dynamic environment where the passengers frequently change their locations. Additionally, the scheme mitigates the sudden failure of the Fog devices. We employ a game-theoretic approach to determine optimal fractions of a task associated with the passengers to be offloaded among the Fog devices and Cloud. It also supports the reallocation of the allocated fractions of the task. This offloading and reallocation of tasks ensures the execution within a given time constraint and requires minimal execution cost. We also prove the existence of near Nash equilibrium for the allocated fractions of the task on Fog devices. Further, this work covers different possibilities of dependencies among the fractions of the task and corresponding utilities of Fog devices in the dynamic environment. Finally, we present the empirical and real-world evaluations to verify the effectiveness of the proposed scheme in terms of the number of Fog devices, the deadline of the task, and game parameters.

An important research issue in smart metering is to correctly transfer the smart meter readings from consumers to the operator within the given time period by consuming minimum energy. In this paper, we propose an energy efficient smart metering system using Edge computing in Long Range (LoRa). We assume that all appliances in a house are connected to a smart meter that is affixed with Edge device and LoRa node for processing and transferring the processed smart meter readings, respectively. The energy consumption of the appliances can be represented as an energy multivariate time series. The system first proposes a deep learning based compression-decompression model for reducing the size of the energy time series at the Edge devices. Next, it formulates an optimization problem for finding the suitable compressed energy time series to reduce the energy consumption and delay of the system. Finally, the system presents an algorithm for selecting the suitable spreading factors to transfer the compressed time series to the operator in the given time. Our simulation and prototype results demonstrate the impact of the parameters of the compression model, network, and the number of smart meters and appliances on delay, energy consumption, and accuracy of the system.

In this paper, we develop techniques that create smart space in an efficient manner, wherein the efficiency is defined in terms of all-together: energy, security, delay, and cost. We design an energy-efficient smart space system using the Long-Range (LoRa) network. The system consists of various sensors that generate sensory data represented as Multi-dimensional Time Series (MTS). The sensors are connected with an Edge device and LoRa node for processing and transferring the MTS, respectively. The system first proposes a deep learning-based compression-decompression model for reducing the size of MTS at the Edge devices. Next, it uses game theory for finding a minimum-cost security mechanism to facilitate the secure transmission of MTS. Then, we formulate an optimization problem to obtain a suitable compression ratio and security mechanism to reduce the system's energy consumption, delay, and security cost. Finally, the system presents an algorithm for selecting the suitable spreading factors to transfer the compressed and secured MTS to the application server in the given time period with desired accuracy. We evaluate the proposed system over the simulation platform and demonstrate the impact of the parameters of the compression model, network, security mechanism, and the number of sensors on energy consumption, delay, cost, and system accuracy.

With the expeditious evolution in technology, recent years have witnessed significant growth in passenger assistance applications in the transportation system. Such applications have varying demands for resources and quality of services. This paper presents a Fog computing based transportation system. The system uses multiple Fog devices to provide assistance to the passengers. The passengers and vehicles work as end-users and the Edge devices in the system, respectively. A passenger generates a task and using the Edge device forwards it to the Fog devices for further processing. Selected Fog devices parallel process the fraction of the task, so that the complete task processes within the given time constraint. We use the gamma function based reputation model of Fog devices, which provides the confidence to complete a given task successfully. We present a Knapsack based task offloading algorithm, which helps to fully utilize the resources of the Fog devices. We also present a competitive game model and near Nash Equilibrium solution for estimating the optimal value of the fraction of the task process at Fog devices. Finally, we develop a prototype and present results to investigate the performance of the propose system.

Coronavirus outbreak is one of the challenging pandemics for the entire human population on Earth. Techniques, such as the isolation of infected people and maintaining social distancing, are the only preventive measures against the pandemic. The actual estimation of the number of infected peoples with limited data is an indeterminate problem faced by data scientists. There are several techniques in the existing literature, including reproduction number and case fatality rate, for predicting the duration of a pandemic and infectious population. This article presents a case study of different techniques for analyzing, modeling, and representing the data associated with a pandemic such as COVID-19. We further propose an algorithm for estimating infection transmission states in a particular area. This work also presents an algorithm for estimating end time of a pandemic from the susceptible infectious and recovered model. Finally, this article presents the empirical and data analysis to study the impact of transmission probability, rate of contact, infectious, and susceptible population on the pandemic spread.

Recent years have witnessed unprecedented growth in sensors-based indoor activity recognition. Further, a significant improvement in recognition performance of indoor activities is observed by incorporating Deep Neural Network (DNN) model. In this paper, we propose knowledge distillation based economic and efficient indoor activity recognition approach for low-cost resource constraint devices. Here, we adopt knowledge from teacher and trainee (cumbersome DNN models) for training student (compressed DNN model). Initially, student and trainee both are beginner and trainee helps the student in learning from the teacher. The student, after certain steps, is mature enough for directly learning from the teacher. We introduce an early halting mechanism for simultaneously reducing floating-point operations and training time of the student model.