Reliable bearing-failure prognosis in induction machines remains a critical research challenge, as it directly impacts system availability, maintenance efficiency, and overall operational safety. To address this challenge, it is essential to develop an online prognostic system capable of continuously assessing bearing health and predicting future failures in real time. This paper proposes a novel unsupervised data-driven prognostic framework for induction machine bearings that integrates advanced signal processing techniques for the preprocessing step, data clustering to construct bearing health stage (HS), artificial neural network (ANN) forecasting using a designed health indicator (HI) based on the latest historical observations, and a fine-tuning model to improve the estimation of remaining useful life (RUL) for induction machine bearings using vibration and temperature signals provided by the PRONOSTIA and NASA-IMS experimentation platform. The results show that the proposed approach is an effective way for bearing RUL estimation.

With growing environmental challenges and rising demands for sustainability, researchers are increasingly focusing on energy-efficient scheduling methods. Time-of-Use (ToU) electricity tariffs play a crucial role in balancing supply and demand, particularly in energy-intensive industries. This paper addresses the Permutation Flow Shop Scheduling problem (PFSP) with the objective of minimizing the total energy cost (T EC) under ToU tariffs, while constraining the schedule’s makespan within a specified time horizon. A Q-learning-based Genetic Algorithm (QL-GA) is proposed, which leverages Q-learning (QL) for dynamic parameter control. The proposed algorithm is evaluated against a classical GA on 25 instances from the challenging VRF benchmark (Vallada et al., 2015), which include both small and large instances with up to 200 jobs and 40 machines. The results demonstrate that integrating Q-learning significantly reduces the TEC and consistently outperforms the classical GA. These outcomes underscore the significance of incorporating reinforcement learning techniques into metaheuristic frameworks. This work contributes to advancing sustainable manufacturing through intelligent scheduling optimization.

In light of growing environmental concerns and rising energy costs, the manufacturing industry is increasingly focusing on energy-efficient scheduling. This paper addresses the Non-Permutation Flow Shop Scheduling (NPFS) problem under Time-of-Use (TOU) pricing, where job sequences may vary across machines. The objective is to minimize the Total Energy Cost (TEC) while maintaining a constraint on the makespan to remain within the time horizon, thereby balancing production efficiency and sustainability. To achieve this, VNS-QL, a hybrid algorithm, is proposed, which uses Q-learning for operator selection within the Variable Neighborhood Search framework. The algorithm is evaluated through experiments on 20 challenging instances adapted from Vallada et al., 2015 and extended to incorporate TEC under TOU pricing. Results show that VNS-QL consistently outperforms VNS in reducing TEC. These findings highlight the value of integrating reinforcement learning into metaheuristic frameworks to guide the search more efficiently, achieving notable energy savings

Fully Homomorphic Encryption (FHE) enables meaningful computations directly on encrypted data, without requiring decryption. Despite its transformative potential, FHE is still computationally expensive. Moreover, writing efficient code for FHE is an error-prone, tedious, and time-consuming task that requires cryptographic expertise. Programmers may not be aware of all possible optimizations, and some optimizations require substantial effort. In this paper, we propose CHEHAB, a compiler that automatically vectorizes scalar code, optimizes it, and generates highly efficient code for FHE. CHEHAB supports the optimization of both structured and unstructured code. It takes as input a program expressed in the CHEHAB domain-specific language (embedded in C++) and uses a Term Rewriting System (TRS) to simplify it. CHEHAB targets two problems: 1) the vectorization of scalar code and 2) the reduction of instruction execution latency and noise growth. It solves these two problems using a TRS that leverages equality saturation to vectorize code, reduce the number of instructions, and reduce the circuit depth. Compared to related work, the proposed approach not only scales better to larger arithmetic circuits but also generates faster code. We evaluate CHEHAB on a set of kernels and compare it to Coyote, a state-of-the-art vectorizing compiler for FHE. We show that CHEHAB generates code that is, on average, 7.38 Chi faster in execution, 2.49 Chi lower in accumulated noise, and 251 Chi faster in compilation.

Deep learning models, particularly those used for im-age classification, have become pervasive in today's technological landscape. However, the increasing demand for privacy, espe-cially when sharing data with external parties, poses significant challenges. Homomorphic encryption (HE) offers a compelling solution by enabling computations on encrypted data without compromising privacy. Despite its potential, the widespread adoption of HE remains limited due to its computational complexity. In this work, we propose an optimized approach to improve the efficiency of a fundamental building block in image classification models: the convolution operation. Our approach is based on the observation that multiple kernel evaluations in a convolution layer can be performed in parallel, provided that their receptive fields do not overlap. Using this property, we reorganize convolution computations into independent groups, significantly reducing redundant operations. This grouping reduces the number of multiplications by up to 75% compared to standard encrypted convolution techniques. Additionally, the approach eliminates the need for costly preprocessing, enabling direct processing of encrypted inputs. To further accelerate homomorphic operations, we leverage a GPU-accelerated HE library, enabling faster op-erations. These contributions make encrypted convolution signifi-cantly more practical for real-world deep learning applications.

Community detection in social networks plays a crucial role in understanding the underlying structure and dynamics of complex systems. In this paper, we present a novel approach for enhancing community detection by refining community boundaries within the framework of Bee Swarm Optimization (BSO). Our method focuses on identifying and contracting the contour of communities, where nodes exhibit greater external degrees than internal degrees, thus enhancing the coherence and granularity of detected communities. Through BSO, we iteratively explore the solution space, employing a contour-based contraction operation to refine community structures. Experimental evaluations on fifteen real-word networks demonstrate the effectiveness of our approach, revealing significant improvements in modularity compared to traditional and state-of-the-art methods.

While polyhedral compilers have shown success in implementing advanced code transformations, they still have challenges in selecting the most profitable transformations that lead to the best speedups. This has motivated the use of machine learning to build cost models to guide the search for polyhedral optimizations. State-of-the-art polyhedral compilers have demonstrated a viable proof-of-concept of this approach. While such a proof-of-concept has shown promise, it still has significant limitations. State-of-the-art polyhedral compilers that use a deep-learning cost model only support a small subset of affine transformations, limiting their ability to apply complex code transformations. They also only support simple programs that have a single loop nest and a rectangular iteration domain, limiting their applicability to many programs. These limitations significantly impact the generality of such compilers and autoschedulers and put into question the whole approach. In this paper, we introduce LOOPer, the first polyhedral autoscheduler that uses a deep-learning based cost model and covers a large set of affine transformations and programs. It supports the exploration of a large set of affine transformations, allowing the application of complex sequences of polyhedral transformations. It also supports the optimization of programs with multiple loop nests and with rectangular and non-rectangular iteration domains, allowing the optimization of an extensive set of programs. We implement and evaluate LOOPer and show that it achieves speedups over the state-of-the-art. On the Polybench benchmark, LOOPer achieves a geometric mean speedup of 1.59x over Tiramisu. LOOPer also achieves competitive speedups with a geometric mean speedup of 1.34x over Pluto, a state-of-the-art polyhedral compiler that does not use a machine-learning based cost model.

With its various real-life applications, predicting community evolution is a challenging task in the field of social network analysis. In this paper, we analyze communities’ evolution prediction accuracy in dynamic social networks. The proposed approach combines several key concepts of the process: (1) a tailored network splitting that results in snapshots of different periods rather than a static one, (2) an event detection method for simultaneously detecting and tracking community structures in dynamic social networks, and (3) the change rates of communities’ features that characterize them over time instead of absolute values of features. Our experiments on four real-world social networks confirm that community evolution prediction can be achieved with a very high accuracy by using both tailored network splitting as a first step of prediction process and change rates of features.

Dynamic community detection is a critical challenge in modern social network analysis and artificial intelligence research. The goal is to improve the quality of community structures as networks evolve while minimizing discrepancies between consecutive community structures. This study introduces Transferring Cliques and Quasi-cliques Knowledge with Multi-Objective Bee Swarm Optimization, referred to as TCK-MBSO, a novel approach aimed at enhancing dynamic community detection performance. The core idea of TCK-MBSO is to transfer valuable clique and quasi-clique information from previous community structures to the current detection task, leveraging this knowledge to improve both accuracy and consistency over time. By embedding this transferred knowledge into a multi-objective bee swarm optimization framework, TCK-MBSO significantly enhances the detection of dynamic communities. Experimental results consistently demonstrate that TCK-MBSO outperforms state-of-the-art algorithms across various test scenarios, delivering superior results in terms of both community structure quality and continuity between snapshots.

Flow Shop Scheduling Problems (FSP) represent a challenging class of combinatorial optimization tasks with significant relevance across diverse domains ranging from manufacturing and logistics to service operations. Recent advances in quantum annealing and quantum-inspired approaches have opened avenues for tackling such problems via Quadratic Unconstrained Binary Optimization models. However, their performance depends strongly on how the problem is encoded as a QUBO. In this study, we perform a comparative evaluation of manually formulated QUBOs versus those automatically produced by the AutoQUBO framework. We test both formulations on D-Wave's hybrid and qbsolv solvers and the InfinityQ solver, exploring the effect of QUBO-solver-specific behaviors and sensitivity, and fine-timing strategies on the quality of the solution. While carefully crafted manual QUBO5 typically yield stronger solution fidelity, they demand greater domain expertise and computational overhead. In contrast, automated approaches greatly expedite modeling at the cost of certain performance trade-offs. We further observe that solver fine-tuning and hardware selection substantially shape outcomes, with certain solvers favoring domain-tailored encodings. These insights demonstrate the necessity of solver-aware QUBO design, offering practical guidance on QUBO construction and solver choice to solve the FSP using quantum and quantum-inspired hardware.

Purpose This paper aims to introduce a novel modularity-based framework, Com_Tracker, designed to detect and track community structures in dynamic social networks without recomputing them from scratch at each snapshot. Despite extensive research in this area, existing approaches either require repetitive computations or fail to capture key community behavioral events, both of which limit the ability to generate timely and actionable insights. Efficiently tracking community structures is crucial for real-time decision-making in rapidly evolving networks, while capturing behavioral events is necessary for understanding deeper community dynamics. This study addresses these limitations by proposing a more efficient and adaptive solution. It aims to answer the following questions: How can we efficiently track community structures without recomputation? How can we detect significant community events over time? Design/methodology/approach Com_Tracker models dynamic social networks as a sequence of snapshots. First, it detects the community structure of the initial snapshot using a static community detection algorithm. Then, for each subsequent time step, Com_Tracker updates the community structure based on the previous snapshot, allowing it to track communities and detect their changes over time. The locus-based adjacency encoding scheme is adopted, and Pearson’s correlation guides the construction of neighboring solutions. Findings Experiments conducted on various networks demonstrate that Com_Tracker effectively detects community structures and tracks their evolution in dynamic social networks. The results highlight its potential for real-time tracking and provide promising performance outcomes. Practical implications Com_Tracker offers valuable insights into community evolution, helping practitioners across fields such as resource management, public security, marketing and public health. By understanding how communities evolve, decision-makers can better allocate resources, enhance targeted strategies and predict future community behaviors, improving overall responsiveness to changes in network dynamics. Originality/value Com_Tracker addresses critical gaps in existing research by combining the strengths of modularity maximization with efficient tracking of community changes. Unlike previous methods that either recompute structures or fail to capture behavioral events, Com_Tracker provides an incremental, adaptive framework capable of detecting both community evolution and behavioral changes, enhancing real-world applicability in dynamic environments.

Fingerprint-based biometric systems are widely used because of their advantages against conventional authentication systems based on passwords and tokens. However, a major limitation is that individuals’ fingerprint information cannot be easily changed if compromised. The fuzzy vault is a promising technique that secures fingerprint data by generating a set of data from the fingerprint using an injective function, preventing the original fingerprint from being regenerated. Nevertheless, the fingerprint fuzzy vault is computationally intensive and requires substantial memory resources. We propose enhancing the performance of fingerprint fuzzy vaults and reducing resource consumption using a new chaff point generation technique based on square boundaries and composite representation. We conducted integration testing along with detailed benchmarking of the fingerprint fuzzy vault using square-boundary generation against other techniques proposed in the literature for each stage. The experiments demonstrate that our proposal yields relatively better results in terms of False Rejection Rate, False Acceptance Rate, computational time, the number of chaff points generated, and memory usage.

Abstract Due to the sophistication of recent malware, classical detection approaches are becoming obsolete. Machine Learning for malware detection has emerged as a new trend and is becoming increasingly effective. Indeed, malware generate a tremendous amount of data that should be analyzed and used to detect them. The aim of this paper is to propose a Machine Learning approach to detect both recent and old malware by converting them into images. This approach, which consists of two phases, is based on Transfer Learning through the use of Convolutional Neural Networks (CNN) that extract features from malware images. These features are used to determine the maliciousness of a particular file. We define six strategies, each one is a combination of two image types (Grayscale and Color) and three CNN architectures (VGG, ResNet and Inception). Experimental evaluation has been done to test these six strategies. The strategy that fulfills the most testing goals is Grayscale + ResNet with a testing accuracy of 90.08\%. Even if the first results are promising, the future work is to automate the fine-tuning of the parameters to go through all possible values and obtain the best ones.

The introduction of 6G technology into the Internet of Vehicles (IoV) promises to revolutionize connectivity with ultra-high data rates and seamless network coverage. However, this technological leap also brings significant challenges, particularly for the dynamic and diverse IoV landscape, which must meet the rigorous reliability and security requirements of 6G networks. Furthermore, integrating 6G will likely increase the IoV’s susceptibility to a spectrum of emerging cyber threats. Therefore, it is crucial for security mechanisms to dynamically adapt and learn new attack patterns, keeping pace with the rapid evolution and diversification of these threats - a capability currently lacking in existing systems. This paper presents a novel intrusion detection system leveraging the paradigm of life-long (or continual) learning. Our methodology combines class-incremental learning with federated learning, an approach ideally suited to the distributed nature of the IoV. This strategy effectively harnesses the collective intelligence of Connected and Automated Vehicles (CAVs) and edge computing capabilities to train the detection system. To the best of our knowledge, this study is the first to synergize class-incremental learning with federated learning specifically for cyber attack detection. Through comprehensive experiments on a recent network traffic dataset, our system has exhibited a robust adaptability in learning new cyber attack patterns, while effectively retaining knowledge of previously encountered ones. Additionally, it has proven to maintain high accuracy and a low false positive rate.

In this work, we propose StreamDD, a data streams clustering algorithm that explicitly detects concept drifts. We will mainly focus on Sudden and Incremental concept drifts. StreamDD is based on CluStream and Page-Hinkley test. Its effectiveness and robustness are evaluated through a series of experiments on generated data streams for which we simulate drifts ( Incremental, Sudden and Mixed ). The comparison of StreamDD with DenStream, a density based stream clustering algorithm, shows that our algorithm ensures good quality clustering while being able to detect concept drifts.

Providing high-quality predictions is no longer the sole goal for neural networks. As we live in an increasingly interconnected world, these models need to match the constraints of resource-limited devices powering the Internet of Things (IoT) and embedded systems. Moreover, in the era of climate change, reducing the carbon footprint of neural networks is a critical step for green artificial intelligence, which is no longer an aspiration but a major need. Enhancing the energy efficiency of neural networks, in both training and inference phases, became a predominant research topic in the field. Training optimization has grown in interest recently but remains challenging, as it involves changes in the learning procedure that can impact the prediction quality significantly. This article presents a study on the most popular techniques aiming to reduce the energy consumption of neural networks’ training. We first propose a classification of the methods before discussing and comparing the different categories. In addition, we outline some energy measurement techniques. We discuss the limitations identified during our study as well as some interesting directions, such as neuromorphic and reservoir computing (RC).

Monitoring pollution plumes is a key issue, given the harmful effects they cause. The dynamic of these plumes, which may be important due to meteorological conditions, makes their study difficult. Real-time monitoring in order to obtain an accurate mapping of the pollution dispersion is helpful and valuable to mitigate risks. In this work, we consider a fleet of cooperative drones carrying pollution sensors and operating in order to assess a pollution plume. The latter is assumed to follow a Gaussian Process (GP) with varying parameters. For this use case, we propose an efficient approach to characterize spatially and temporarily the plume while optimizing the path planning of drones. In our approach, drones are guided by a Deep Reinforcement Learning (DRL) model called Categorical Deep Q-Network (Categorical DQN) to maximize the plume coverage while considering budget constraints. Specifically, we develop a scalable Independent Q-Learning (IQL) scheme that shares team rewards based on each drone’s deployment relevance and therefore ensures cooperation. We evaluate the performance of the plume parameter estimation as well as the maps generated by the GP regression. By testing our framework on several plume scenarios, we show that it offers good results in terms of both estimation quality and run-time efficiency.

Association rule mining (ARM) is one of the most popular tasks in the field of data mining, very useful for decision-making. It is an NP-hard problem for which Genetic algorithms have been widely used. This is due to the obtained competitive results. However, their main drawback is the fitness computation which is time-consuming, especially when working with huge data. To overcome this problem, we propose an offline approach in which we substitute the GA’s fitness computation with a Machine Learning model. The latter will predict the quality of the different generated solutions during the search process. The performed tests on several well-known datasets of different sizes show the effectiveness of our approach.

In this paper, we analyze the results of two hyper-heuristics HHGA and HHabs that generate per-instances genetic algorithms for the permutation flow shop problem. They are competitive with literature approaches for most of instances of the benchmark of Taillard. Nevertheless, they are not effective enough for some difficult instances. For this purpose, we propose a workflow to analyse GAs configurations and their results in order to detect which components influence the most on generated GAs quality in order to enhance the quality of these hyper-heuristics.

In this paper, we address the energy efficiency of neural networks training through data selection techniques. We first study the impact of a random data selection approach that renews the selected examples periodically during training. We find that random selection should be considered as a serious option as it allows high energy gains with small accuracy losses. Unexpectedly, it even outperforms a more elaborate approach in some cases.Our study of the random approach conducted us to observe that low selectivity ratios allow important energy savings, but also cause a significant accuracy decrease. To mitigate the effect of such ratios on the prediction quality, we propose to use a dynamic selectivity ratio with a decreasing schedule, that can be integrated to any selection approach. Our first results show that using such a schedule provides around 60% energy gains on the CIFAR-10 dataset with less than 1% accuracy decrease. It also improves the convergence when compared to a fixed ratio.