International audience

International audience

International audience

International audience

Federated Learning (FL) is emerging as a crucial approach to enhance data privacy and security, particularly in smart buildings and Internet of Things (IoT) ecosystems. By distributing learning across multiple clients, FL minimizes the need for centralized data transfers. This decentralized approach allows clients to collaboratively improve machine learning models without sharing raw data, and only their model updates are sent to a central server for aggregation. However, the problem with the existing aggregation approaches is randomizing and fixing the choice of participating clients during the FL process without evaluating the quality and potential anomalies in individual client model updates during training rounds, which can impact the aggregation and the global model performance. Therefore, we introduce a novel dynamic client selection approach called FedCSA, which selects clients using a scoring mechanism that prioritizes model quality and anomaly detection. Clients with scores above a threshold are chosen, and any client not selected for several consecutive cycles is flagged as malicious and removed. This ensures bad or malfunctioning clients are secluded and not selected during the remaining training rounds. Simulation results using smart building datasets demonstrate superior global per-formance compared to other client selection methods, including loss, SMAPE, RMSE, and MAE, across varying client numbers. This shows the scalability and consistency of our method for large-scale FL tasks with IoT time-series data.

Network Intrusion Detection Systems (NIDS) are mechanisms designed to improve security by monitoring networks for signs of potential intrusions. While data-driven deep learning-based NIDSs have been popular for their superior performance, they are limited by their reliance on large amounts of data, often processed in a centralized manner. Federated Learning (FL) has thus emerged as a distributed paradigm to preserve privacy and data confidentiality, reduce communication costs, and promote collaborative learning. However, FL solutions require a high degree of generalization and adaptation to data and system heterogeneity. In this paper, we introduce a new approach to enhance the generalization of deep learning models in FL-based NIDS by integrating centrality measures. These centrality measures assess the importance of nodes in a cyber-physical system, providing valuable insights into network structures. By adopting these measures within the graph constructed from source and destination devices of network flows, we aim to enhance the model's understanding of how network dynamics correlate with intrusion patterns. For our experiments, we used two public datasets: CIC-IDS-2017 and CIC-ToN-IoT. To reflect real-world network variability, we utilized a realistic federated learning setup by distributing distinct parts of the datasets among FL clients. Our approach demonstrates an improvement of over 6 % in F1-score with the use of centrality measures, surpassing the traditional baseline approach. Our findings underscore the effectiveness of integrating centrality measures in FL-based NIDS, offering enhanced intrusion detection capabilities in heterogeneous network environments.

Energy consumption forecasting is of paramount importance in achieving energy conservation goals. While numerous approaches have been developed to optimize building energy usage, predictive analytics stands out as a cornerstone tool for informed decision-making. Deep learning models have gained popularity for forecasting energy consumption in smart buildings. These models leverage a variety of techniques, including loss functions, activation functions, and optimizers, to enhance training effectiveness. However, the commonly used Mean Squared Error (MSE) as a loss function has a notable drawback as it treats overestimations and underestimations equally. In this study, we propose a novel Hybrid Loss Function (HLF) tailored to address this limitation. The HLF penalizes the model more for underestimating energy consumption during abnormal seasons while maintaining its ability to accurately predict actual consumption, particularly under normal operating conditions. Through extensive simulations, our findings demonstrate that our proposed approach outperforms existing methods in the literature, providing exceptionally accurate and robust forecasts of energy consumption.

In the ever-evolving realm of federated learning (FL), the question of data worth resonates with newfound urgency across organizations and individuals. In the dynamic FL ecosystem, where data resides across distributed nodes, evaluating the value of each client’s data is paramount. The evaluation mechanism helps to understand individual contributions to the overall process and incentivizes the best contributors, thereby ensuring the sustainability of federated training. Drawing inspiration from cooperative game theory approaches, we harness the Shapley Value (SV)—a well-established measure of value—to address this challenge. Despite offering valuable insights, the computation of the Shapley Value often entails exponential time complexity. In our study, we propose an Enhanced Guided Truncation Gradient Shapley algorithm, precisely tailored for efficient SV approximation in FL settings. Specifically, our approach comprises two pivotal enhancements for the GTG-Shapley method. First, we optimize the client sampling policy to generate representative permutations. Second, we employ an order-reversed marginal utility function based on the Monte-Carlo estimation for SV calculation. Through empirical experiments, we demonstrate the superior performance of EGTG-Shapley compared to the conventional GTG-Shapley method, showcasing significant efficiency gains in FL contexts.

The heterogeneity induced by the federated edge learning execution environment poses many performance challenges. Indeed, a balance between efficient resource usage and inference accuracy must be found. Our work therefore aims at characterizing the hyperparameter influence by creating a variety of simulated execution circumstances. We designed an experimentation platform to simulate the execution of a typical image recognition training workload to highlight tweaking opportunities. We particularly focus on participant selection as an important performance lever. Thus, our benchmarks vary the number of clients participating in the federated edge learning process within i.i.d. and non-i.i.d. environments, while illustrating real-world configurations based on heterogeneous edge systems. We identify computational efficiency facets in federated edge learning and propose a taxonomic methodology to approach the study. We demonstrate the impact of the number of clients selected to participate in the global model update of federated edge learning on the overall system computational efficiency in challenging environments. Thus, we propose an optimization formula to meet computational efficiency and accurate models in challenging federated edge learning environments.

Federated Learning (FL) is rapidly growing in popularity as a decentralized approach and is being adopted in smart building systems and energy forecasting without accessing sensitive data. Specifically, clients train their models using their own data. After that, only their model parameters are sent to the central server, which aggregates them by averaging the weights and then sends back the newly formed model to each client. However, challenges arise when dealing with heterogeneous multivariate time-series data with different distributions. This leads to higher-performing clients contributing to the global update more than the others, and slower convergence where the global model takes more time to generalize across the clients. In this paper, we propose an enhanced aggregation approach, where the server sorts clients’ models based on their local training losses before swapping them all consecutively according to the best and worst-performing ones. Our proposed approach is applied to a smart building dataset and compared with two other FL approaches from the literature. Our simulation results demonstrate improved forecasting precision for each client and faster convergence. Moreover, we optimized the global model’s evaluation error scores and overall loss, reduced the communication rounds required for convergence, and ensured less bias and more fairness between clients during each training cycle.

Wireless Sensor Networks (WSNs) represent a key component in smart building systems. An efficient WSN deployment involves selecting the most appropriate positions within the building to place sensors in order to maximize coverage and minimize the deployment cost. This paper proposes a novel approach called the Hybrid Binary Grey Wolf Optimiser (HBGWO) to automate the WSN deployment in indoor environments. The proposed approach integrates the Building Information Modeling (BIM) database to accurately model the physical layout and structural characteristics of the deployment area. Furthermore, a Steiner Tree-based heuristic has been developed to reduce the number of active sensors while preserving the network coverage. Experimental results demonstrate the efficiency and superiority of the HBGWO approach compared to existing methods in literature in terms of network coverage and deployment cost under the connectivity constraint.

Finance, supply chain, healthcare, and energy have an increasing demand for secure transactions and data exchange. Permissioned blockchains fulfilled this need thanks to the consensus protocol that ensures that participants agree on a common value. One of the most widely used protocols in private blockchains is the Practical Byzantine Fault Tolerance (PBFT) which tolerates up to one-third Byzantine nodes, performs within partially synchronous systems and has a superior throughput compared to other protocols. It has, however, an important bandwidth consumption: 2N(N-1) messages are exchanged in a system composed of N nodes to validate only one block. It is possible to reduce the number of consensus participants by restricting the validation process to nodes that have demonstrated high levels of security, rapidity, and availability. In this paper, we propose the first database that traces the behavior of nodes within a system that performs PBFT consensus. It reflects their level of security, rapidity and availability throughout the consensus. We first investigate different Single-Task Learning techniques to classify the nodes within our dataset. Then, using Multi-Task learning techniques, the results are way more interesting with classification accuracies over 98%. Integrating nodes classification as a preliminary step to the PBFT protocol optimizes the consensus. In the best cases, it is able to reduce the latency by up to 94% and the communication traffic by up to 99%.

The emerging integration of Internet of Things (IoT) and AI has unlocked numerous opportunities for innovation across diverse industries. However, growing privacy concerns and data isolation issues have inhibited this promising advancement. Unfortunately, traditional centralized Machine Learning (ML) methods have demonstrated their limitations in addressing these hurdles. In response to this ever-evolving landscape, Federated Learning (FL) has surfaced as a cutting-edge ML paradigm, enabling collaborative training across decentralized devices. FL allows users to jointly construct AI models without sharing their local raw data, ensuring data privacy, network scalability, and minimal data transfer. One essential aspect of FL revolves around proficient knowledge aggregation within a heterogeneous environment. Yet, the inherent characteristics of FL have amplified the complexity of its practical implementation compared to centralized ML. This survey delves into three prominent clusters of FL research contributions: personalization, optimization, and robustness. The objective is to provide a well-structured and fine-grained classification scheme related to these research areas through a unique methodology for selecting related work. Unlike other survey papers, we employed a hybrid approach that amalgamates bibliometric analysis and systematic scrutinizing to find the most influential work in the literature. Therefore, we examine challenges and contemporary techniques related to heterogeneity, efficiency, security, and privacy. Another valuable asset of this study is its comprehensive coverage of FL aggregation strategies, encompassing architectural features, synchronization methods, and several federation motivations. To further enrich our investigation, we provide practical insights into evaluating novel FL proposals and conduct experiments to assess and compare aggregation methods under IID and non-IID data distributions. Finally, we present a compelling set of research avenues that call for further exploration to open up a treasure of advancement.

The wireless sensor deployment problem is one of the major issues in wireless sensor networks (WSNs). It involves designing the optimal network topology within the deployment area in order to maximize network coverage and lifetime and minimize cost and energy consumption under the connectivity constraint. The WSN deployment problem is a challenging NP-hard combinatorial optimization problem due to a number of factors, including the size and the type of the deployment area, the number of obstacles, and the number of objectives to optimize. Consequently, metaheuristics are assumed to be the most efficient methods to compute the deployment scheme in a reasonable amount of time. In this paper, several well-known metaheuristics have been tested on the problem of WSN deployment in indoor environments. The problem has been formulated as a constrained single objective optimization problem, and the performance of the selected algorithms has been evaluated through experimentation on a set of ten representative indoor architectural scenarios with varying dimensions and obstacles.

Wireless sensor network connectivity is a crucial parameter since it keeps the network operative. Network connectivity may be lost due to a variety of factors, such as energy depletion and sensor node failure. Therefore, the network will be partitioned into a set of disjoint sets, resulting in a loss of data collected by isolated sets. In this paper, we address the problem of connectivity repair for stationary sensor networks (WSNs) in case of multiple disjoint partitions. We propose two heuristics based on Dijkstra algorithm and minimum Steiner tree respectively, to deploy the minimum number of additional nodes while preserving the initial topology. For the two heuristics, a procedure is executed in the first stage to merge disjoint sets having a shared zone in their neighboring deployment zones to reduce the complexity of the solution. The first heuristic is adapted to free-obstacle areas and areas with few obstacles. It connects the less distant segments using Dijkstra algorithm. The second heuristic is rather appropriate for areas with opaque obstacles. Simulation experiments validate the effectiveness of the proposed methods compared to existing approaches.

As the world’s economy and urbanization develop rapidly, energy shortages and pollution are becoming major challenges. In general, buildings are responsible for approximately 40% of global energy consumption. To combat this challenge, the adoption of intelligent buildings is strongly recommended. Therefore, through the data collected by the smart buildings, effective solutions should be incorporated to limit their impact on the environment; Machine learning (ML) has proved great success in sensor-based Energy consumption. Trapping in 10-cal optima and slow convergence are the main difficulties of the Backpropagation (BP) learning algorithm. This has an impact on the neural network’s performance. To recover the drawback, this paper proposes a hybrid protocol FedLM-PSO that combines Particle Swarm optimization (PSO) and Levenberg Marquardt (LM) to train MLP models in a Federated Learning environment to find the near-optimal configurations for FL. In addition, FedLM-PSO evolves the way clients upload data to servers and reduces the amount of data sent, which enhances bandwidth consumption. According to the results, the FedLM-PSO is more accurate and requires fewer rounds of communication than FedAVG.

As smart cities become more decentralized, the need for reliable and secure cyber-physical systems (CPS) that guarantee safe interactions and secure data storage without loss of privacy is continuously increasing. Blockchain is a rapidly emerging technology in this domain. It demonstrated effectiveness thanks to the cryptographic mechanisms it utilizes and to its immutability. Private blockchains are the most suited to applications that require privacy and confidentiality when data is very sensitive. In this case, the most commonly used consensus protocol is Practical Byzantine Fault Tolerance (PBFT). However, PBFT requires the participation of all nodes in the consensus process, which increases bandwidth consumption and consensus delay significantly. In this paper, we propose an adaptive PBFT protocol called APBFT that optimizes the number of nodes participating in the consensus based on their response time and credibility. Therefore, we reduce the amount of communication and the response delays. We maintain the asynchrony of the algorithm so that it remains resilient to DoS attacks. The simulation results show that our algorithm outperforms the original PBFT in terms of delays and message traffic.

This paper proposes a new mechanism of path selection and admission control for IPTV in IEEE 802.16j simultaneously. The proposed mechanism takes into account some constraints of Quality of Services (QoS) which are required by real-time applications, such as available bandwidth, end-to-end delay, number of hops between MR-BS and SS as well as quality of radio signal. With all these four constraints, selecting the best path becomes a multi-objective optimization problem, since we have two criteria to maximize and two other criteria to minimize. It makes selection process of the optimal path more difficult to find a solution in polynomial time. To solve this multi-constrained problem, the proposed approach applied a cost function which simplifies the multi-objective problem into single objective. This function provides a deterministic solution which takes into account all the above constraints. To evaluate the proposed mechanism, we study the performance of the proposed approach through various simulation scenarios. The results show that the proposed mechanism outperforms other studies.