Federated learning (FL) has emerged as a transformative paradigm enabling collaborative machine learning without centralizing data, preserving client privacy. This is particularly relevant in the context of edge computing, where the proliferation of Internet of Things devices has led to an explosion of data at the network’s edge. These IoT devices, often battery-powered, are limited by their energy capacities, which pose significant challenges for the adoption of FL in such environments. In this paper, we introduce BatteryFL, a novel framework that coordinates battery-aware clients through FL to maximize their contribution to the global model while ensuring a fair distribution of energy consumption across the clients without compromising accuracy. BatteryFL incorporates an innovative data collection algorithm that prioritizes data diversity to minimize battery usage and a sample relevance-based algorithm to select optimal data for training. We also integrate a client selection strategy into the framework to optimize training loss and fairness (based on the battery energy of the clients) simultaneously. Along with a theoretical analysis, we experimentally demonstrate that BatteryFL significantly improves the energy efficiency of FL, prolonging the data collection and the contributions of the clients.

Smart farming systems encounter significant challenges, including limited resources, the need for data privacy, and poor connectivity in rural areas. To address these issues, we present eEnergy-Split, an energy-efficient framework that utilizes split learning (SL) to enable collaborative model training without direct data sharing or heavy computation on edge devices. By distributing the model between edge devices and a central server, eEnergy-Split reduces on-device energy usage by up to 86% compared to federated learning (FL) while safeguarding data privacy. Moreover, SL improves classification accuracy by up to 6.2% over FL on ResNet-18 and by more modest amounts on GoogleNet and MobileNetV2. We propose an optimal edge deployment algorithm and a UAV trajectory planning strategy that solves the Traveling Salesman Problem (TSP) exactly to minimize flight cost and extend and maximize communication rounds. Comprehensive evaluations on agricultural pest datasets reveal that eEnergy-Split lowers UAV energy consumption compared to baseline methods and boosts overall accuracy by up to 17%. Notably, the energy efficiency of SL is shown to be model-dependent—yielding substantial savings in lightweight models like MobileNet, while communication and memory overheads may reduce efficiency gains in deeper networks. These results highlight the potential of combining SL with energy-aware design to deliver a scalable, privacy-preserving solution for resource-constrained smart farming environments.

Federated Learning (FL) is a distributed learning paradigm that leverages the computational strength of local devices to collaboratively train a model. The clients train the local model on their respective devices and submit the weight updates to the server for aggregation. This paradigm allows the clients to experience diverse data without sharing their local data with other participants or the server. However, FL is susceptible to backdoor attackers that deliberately train the model on altered data, essentially trying to get favor on a specific subtask separated from the main task. In this work, we focus on powerful backdoor attackers who play attacks in a distributed manner to strengthen their impact and to escape a strong detection method. We propose a novel defense algorithm against distributed backdoor attacks, which leverages dynamic model clipping and a reputation-based global model update by filtering adversarial update vectors. While requiring minimal changes to the standard FL framework, our algorithm can be used as a plug-in solution. By simulating various forms of backdoor attacks over three benchmark datasets, we find that with a negligible compromise on the overall performance of the model, our algorithm maintains a lower attack success rate and outperforms the prior solutions.

Federated Learning (FL) is a distributed machine learning paradigm facilitating participants to collaboratively train a model without revealing their local data. However, when FL is deployed into the wild, some intelligent clients can deliberately deviate from the standard training process to make the global model inclined toward their local model, thereby prioritizing their local data distribution. We refer to this novel category of misbehaving clients as selfish. In this paper, we propose a Robust aggregation strategy for the FL server to mitigate the effect of Selfishness (in short RFL-Self). RFL-Self incorporates an innovative method to recover (or estimate) the true updates of selfish clients from the received ones, leveraging robust statistics (median of norms) of the updates at every round. By including the recovered updates in aggregation, our strategy offers strong robustness against selfishness. Our experimental results, obtained on MNIST and CIFAR-10 datasets, demonstrate that just 2% of clients behaving selfishly can decrease the accuracy by up to 36%, and RFL-Self can mitigate that effect without degrading the global model performance.

In the last few years, Federated Learning (FL) has received extensive attention from the research community because of its capability for privacy-preserving, collaborative learning from heterogeneous data sources. Most FL studies focus on either average performance improvement or the robustness to attacks, while some attempt to solve both jointly. However, the performance disparities across clients in the presence of attackers have largely been unexplored. In this work, we propose a novel Fair Federated Learning scheme with Attacker Detection capability (abbreviated as FFL+AD) to minimize performance discrepancies across benign participants. FFL+AD enables the server to identify attackers and learn their malign intent (e.g., targeted label) by investigating suspected models via top performers. This two-step detection method helps reduce false positives. Later, we introduce fairness by regularizing the benign clients’ local objectives with a variable boosting parameter that gives more emphasis on low performers in optimization. Under standard assumptions, FFL+AD exhibits a convergence rate similar to FedAvg. Experimental results show that our scheme builds a more fair and more robust model, under label-flipping and backdoor attackers, compared to prior schemes. FFL+AD achieves competitive accuracy even when 40% of the clients are attackers.

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.

In the era of a data-driven society with the ubiquity of Internet of Things (IoT) devices storing large amounts of data localized at different places, distributed learning has gained a lot of traction, however, assuming independent and identically distributed data (iid) across the devices. While relaxing this assumption that anyway does not hold in reality due to the heterogeneous nature of devices, federated learning (FL) has emerged as a privacy-preserving solution to train a collaborative model over non-iid data distributed across a massive number of devices. However, the appearance of malicious devices (attackers), who intend to corrupt the FL model, is inevitable due to unrestricted participation. In this work, we aim to identify such attackers and mitigate their impact on the model, essentially under a setting of bidirectional label flipping attacks with collusion. We propose two graph theoretic algorithms, based on Minimum Spanning Tree and k-Densest graph, by leveraging correlations between local models. Our FL model can nullify the influence of attackers even when they are up to 70% of all the clients whereas prior works could not afford more than 50% of clients as attackers. The effectiveness of our algorithms is ascertained through experiments on two benchmark datasets, namely MNIST and Fashion-MNIST, with overwhelming attackers. We establish the superiority of our algorithms over the existing ones using accuracy, attack success rate, and early detection round.

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.

Network traffic classification has been an interesting topic of research for many years. It plays a crucial role in many network applications including resource allocation, intrusion detection, and quality of service. Network traffic is essentially a sequence or flow of time-stamped packets that are exchanged between two devices. The traffic flow also contains payload data along with information about packet statistics such as size, inter-arrival time, and direction. As these statistics are obtained from the time-stamped packets, they form a Multivariate Time Series (MTS). Such an MTS needs to be classified as early as possible to identify an Internet application associated with the generated traffic flow. In this paper, we propose an Early traffic Flow Classification (EFC) approach for identifying Internet applications using MTS. The approach estimates application-wise minimum required packets from the training data by employing k-means clustering and Long Short Term Memory model. We also develop a class forwarding method to utilize correlation that exists among different packet statistics. Additionally, we collect a real-world traffic flow dataset to evaluate the effectiveness of the approach. Experimental results show that EFC approach requires only the first 15 packets of the flow to achieve an accuracy of more than 90%.

Getting real-time information about the operational behavior of industrial or domestic appliances becomes effortless with the availability of sensors. The sensors generate multiple streams of measurements, called as multivariate time series, corresponding to an operation of the appliance. An appliance can go through various types of faults during its lifetime. Such a fault can be identified by classifying the multivariate time series (MTS), which is generated by the sensors corresponding to this fault. As it is also unfeasible to have prior knowledge about all types of faults, the classification approach should also be able to identify an unseen (unknown) fault using its MTS. In this article, we propose a semantic-information-based early classification approach for MTS. The approach uses a concept of zero-shot learning to classify an unseen fault. This work conducts a case study to evaluate the approach by classifying different faults of a washing machine using sensory data.

Activity classification has been an interesting area of research for many years, to better understand human behavior. Recent advancements in embedded computing systems allowed the emergence of several state-of-art solutions for human activity classification using sensors of a smartphone. The sensors generate temporal sequences of observations for human activity, which is called as Multivariate Time Series (MTS). Current state-of-art solutions for human activity classification suffer from two major limitations: first, the length of testing MTS should be equal to the training MTS and second, the MTS should not have any faulty time series. In real-time applications, it is desirable to classify a human activity using an incomplete MTS as early as possible. In this work, we propose a fault-tolerant early classification of MTS (FECM) approach to address these limitations. FECM builds a set of classification models using MTS training dataset. The approach employs Gaussian Process classifier to estimate minimum required length of time series, which is used to predict a class label of new MTS. Further, FECM uses an Auto Regressive Integrated Moving Average model to identify faulty time series in the new MTS. Finally, we conduct an experiment to evaluate the performance of FECM using accuracy and earliness metrics.

In the era of the Internet of Things (IoT), the sensor-based devices produce the Multivariate Time Series (MTS). A classification approach helps to predict the class label of an incoming MTS. Due to the large dimension and different sampling rate of the sensors in a given MTS, a classifier takes time to predict the class label. Some IoT applications may require early prediction of the class label where the classifier starts the prediction once the minimum number of data points are collected. In this article, we address the problem of early prediction of the class label of an MTS in IoT. This work considers the sensors with different sampling rate to generate the MTS. Each sensor generates a time series (component) of the MTS. We propose a Divide-and-Conquer–based early classification approach for classifying such MTS. The approach constructs an ensemble classifier using a probabilistic classifier and hierarchical clustering. The ensemble classifier employs a Divide-and-Conquer method to handle the different sampling rate components during the prediction of class label. The experimental results show that our approach significantly outperforms the existing approaches on real-world datasets using various evaluation metrics.

An important issue of research in the transportation system is timely classification of the inside-outside environment of the vehicle using sensors. The sensors generate the multivariate time series data, which requires a classification technique to classify it in real-time. Road surface classification is an example, where multivariate time series data can be used for early identification of the type of road surface. The challenge is to maintain the accuracy of the classification using a minimum number of data points of the multivariate time series. This work proposes an early classification approach for multivariate time series with a desired level of accuracy. It is assumed that the number of samples in the time series are not equal for a given period of time due to different type of sensors. Gaussian Process learning method is used to first estimate the minimum required length of the time series which helps to build an ensemble classifier with a desired level of accuracy. The ensemble classifier is used to predict the class label of an incoming multivariate time series. This work demonstrates a road surface classification system using the built ensemble classifier. Finally, the ensemble classifier is also evaluated on the various existing datasets from other domains. The results demonstrate the significance of early classification approach using accuracy, earliness, and confusion matrix, with the minimum required data points.

Classification of Multivariate Time Series (MTS) data has been an important area of research for many years. In time-critical applications, such as health informatics, fire detection, and disaster forecasting, it is desirable to classify the MTS data as early as possible. This work proposes an early classification approach to classify an incoming MTS. The early classification approach helps to predict the class label of an incoming MTS without waiting for the full length. Different from the existing work, this work considers that sampling rate of the sensors which generated the MTS is different. The performance of the approach is evaluated on a publicly available dataset using accuracy, earliness and energy consumption.