Linking APIs to the business functions they implement is crucial for handling software operations, especially during recovery from disasters or outages. In this context, the speed and accuracy of operators in linking them impact response time during mission-critical operation activities. Besides, this linkage is essential to designing preventive actions, such as resilience strategies. Automatic API classification using Large Language Models (LLMs) may simplify and speed up APIs-business function linkage. However, previous studies unveiled the barriers practitioners face when deciding on and adopting LLMs in software engineering (SE) tasks due to a lack of guidance for non-experts. This paper aims to lower barriers to using LLMs by systems operators and site reliability engineers (SREs), focusing on the API classification task in the context of operational activities. Based on three cases from the finance industry, we extracted requirements for LLM usage, and assessed 14 recently released LLMs on this task. Our results demonstrate that LLMs accurately classify APIs using business function targets with an F1–Score of 89.5 for the leading LLM without requiring specific LLM expertise and resource-intensive fine-tuning. Besides, our findings on LLMs’ performance and reliability mark a significant advancement in comparing open and closed-source and general and domain-specific LLMs in an SE classification task. Eventually, our experiments yield practical guidance for implementing LLMs in this context. Artifacts used in and generated by the experiments are publicly available at https://bit.ly/llms4apiclassification.

Abstract Protecting user privacy is essential in machine learning research, especially in the context of data collection. Federated learning (FL), which trains models across decentralized devices without sharing raw data, has emerged as a promising solution. However, FL is still vulnerable to security threats, including inference attacks, which have been underexplored in comparison to poisoning and backdoor attacks that have received more attention in existing research. To address these vulnerabilities, this paper proposes a novel aggregation framework called homomorphic and polymorphic federated learning aggregation of parameters (HP_FLAP). HP_FLAP integrates both homomorphic and polymorphic encryption to enhance the security and privacy of FL. Homomorphic encryption allows the server to perform aggregation on encrypted parameters without decrypting them, ensuring that sensitive information is protected during the aggregation process. Polymorphic encryption further strengthens security by using different encryption keys for each set of parameters, mitigating the risk of system-wide compromise in case a key is leaked. This dual encryption approach effectively counters inference attacks while maintaining robust protections against other security threats. The framework is evaluated using multiple models, including logistic regression, Gaussian Naive Bayes, Stochastic Gradient Descent, and Multi-Layer Perceptron, demonstrating HP_FLAP’s ability to enhance both security and privacy in FL environments.

Smart Grid (SG) technology utilizes advanced network communication and monitoring technologies to manage and regulate electricity generation and transport. However, this increased reliance on technology and connectivity also introduces new vulnerabilities, making SG communication networks susceptible to large-scale attacks. While previous surveys have mainly provided high-level overviews of SG architecture, our analysis goes further by presenting a comprehensive architectural diagram encompassing key SG components and communication links. This holistic view enhances understanding of potential cyber threats and enables systematic cyber risk assessment for SGs. Additionally, we propose a taxonomy of various cyberattack types based on their targets and methods, offering detailed insights into vulnerabilities. Unlike other reviews focused narrowly on protection and detection, our proposed categorization covers all five functions of the National Institute of Standards and Technology cybersecurity framework. This delivers a broad perspective to help organizations implement balanced and robust security. Consequently, we have identified critical research gaps, especially regarding response and recovery mechanisms. This underscores the need for further investigation to bolster SG cybersecurity. These research needs, among others, are highlighted as open issues in our concluding section.

The Internet of Things (IoT) aims for connecting. This assumption brings about several software engineering challenges that constitute a serious obstacle to its wider adoption. The main feature of the IoT is genericity w.r.t the variability of software and hardware technologies. Model-driven engineering (MDE) is a paradigm that advocates using models to address software engineering problems. It can help to meet the genericity of the IoT from a software engineering perspective. Existing MDE approaches for the IoT focus only on modeling the internal behavior of things but lack a comprehensive approach dedicated to network modeling. In the present article, we introduce a network-oriented methodology based on MDE to unify the IoT’s heterogeneous concepts. Fundamentally, we avoid the intrinsic heterogeneity of the IoT by separating the network’s specification (i.e., the things, the communication scheme, and its constraints) from its concrete implementation (i.e., the low-level artifacts, such as source code and documentation). Technically, the methodology relies on a model-based domain-specific language (DSL) and a code generator. The former enables the modeling of the network’s specification, and the latter provides a procedure to generate the low-level artifacts from this specification. Our results show that this methodology makes iot’s software engineering more rigorous, helps prevent bugs earlier, and saves time.

Ambient Assisted Living (AAL) aims at assisting people in their Activities of Daily Living (ADL). We have seen an increased interest in their applicability to the rural seniors who are slowly losing their autonomy due to aging and chronic diseases. By deploying intelligent devices in the environment of an individual performing their ADLs, we can gather data in the form of a time series. One research venue is to seek to use forecasting techniques to discover trends and predict future trends that could be used to analyze the health of these individuals. With the recent advances in computational power new deep learning forecasting algorithms have been developed. In this paper, we compare a univariate one-dimensional CNN model and a LSTM model that performs multi-step forecasting for one week ahead. The novel dataset used comes from a set of activity and health related sensors deployed in a small apartment that uses our previously designed analytics architecture. We compare these to a forecasting baseline strategy. Both deep learning approaches increase the forecasting accuracy significantly.

Microservices Architectures (MSAs) are continuously replacing monolithic systems toward achieving more flexible and maintainable service-oriented software systems. However, the shift toward an MSA also requires a technological and managerial shift for its adopters. Architecting and managing MSAs represent unique challenges, including microservices' identification, interoperability, and reuse. To handle these challenges, we propose an Ontology-driven Conceptual Modelling approach, based on the Ontology of Microservices Architecture Concepts (OMSAC), for modelling and analyzing microservices-based systems. We show, how OMSAC-based conceptual models, stocked in a Stardog triple store, support Stakeholder-specific communication, documentation, and reuse. This paper reports on the application of our approach in three open-source MSA systems with a focus on microservices' discovery based on similarity metrics. Eventually, we compare the extracted similarity metrics derived from the application of machine learning techniques to the OMSAC models with a manual analysis performed by experts.

Ambient Assisted Living (AAL) in general and Activity Recognition (AR) in particular are active fields of research that aim at assisting people in their Activities of Daily Living (ADL). In recent years, we have seen an increased interest in their applicability to the rural seniors who are slowly losing their autonomy due to aging and chronic diseases. One research venue is to aggregate and seek for correlations between the physiological data that serves to monitor the health of the elderly, their ADLs, their movements and any other data that may be collected about their immediate environment. In this paper, we are tackling the possibility of developing a non-intrusive and affordable system based on embedded health, movement, activity and location sensors. Furthermore, we discuss the main concepts behind the creation of a layered, flexible and highly modular architecture that focuses on how the integration of newly combined sensor data can be achieved. Using a mobile phone application prototype, our work has shown that we can integrate two non-invasive technologies that are not necessarily the newest, but the most affordable, scalable and ready to be deployed in real life settings.

Posture recognition is one of the challenging tasks in computer vision. It lays on top of the pose estimation of the different body joints, and can be used in many applications. In the medical field, it can serve to assist patients in rehabilitation. In games it can be an elegant form of computer human interaction. Different Artificial Intelligence techniques were used over the years to precisely output the joint positions of the body from a single or stream of images. One of the great solutions that tacked well the pose estimation challenge is the Kinect camera, however further process is required to create and detect body postures. This article presents a customizable high-level layer that allows its users to easily create and manage body postures in unity3d projects allowing them to focus more on the other aspects of their project. The layer offers two detection methods, both scored more than 95% accuracy in each of the tested postures.

Many models have been proposed to address deep learning problem. Most deep learning models are influenced by presentation order, complex shapes, architecture configuration and learning instability. This paper provides comparative study to deep learning for pattern recognition. Two types of supervised learning techniques were tested which are used for comparison purpose. They correspond to Batch Gradient Descent and Stochastic Gradient Descent. In order to obtain an accurate results with both methods, we used a re-sampling method based on k-fold cross-validation. Experimental Results show that Stochastic Gradient Descent gives good results in comparison to Batch Gradient Descent. The recognition accuracies are seen to improve significantly when Stochastic Gradient Descent is applied for intrusion detection.

Human-Computer Interaction represents an important component in each device designed to be used by humans. Moreover, improving interaction leads to a better user experience and effectiveness of the designed device. One of the most intuitive ways of interaction remains emotions since they allow to understand and even predict the human behavior and react to it. Nevertheless, emotion recognition still challenging since emotions might be complex and subtle. In this paper, we introduce a new hybrid-based approach to identify emotions through facial expressions. We combine two different feature types that are geometric-based (from facial fiducial points) and appearance-based (from Discrete Wavelet Transform coefficients). Each one provides specific information about the six basic emotions to identify. Furthermore, we propose to use a multi-class Support Vector Machine architecture for classification and Extremely Randomized Trees as feature selection technique. Carried experimentation attests to the effectiveness of our approach since it yields 96.11%, 91.79% and 99.05% with three benchmark facial expression datasets namely JAFFE, KDEF and RaFD.

Recommendation technique is a personalized search used to assist a user access information/services that are related to his preferences and interests, or to the preferences and interests of similar users. The main challenge of personalized Information Retrieval is the modeling and the integration of user profiles. In this paper, we propose a generic model of user profiles based on the search history of users delimited by several search sessions. These profiles are based on weighted topical graphs and are integrated into a hybrid data recommendation process. To evaluate the proposed system a prototype is developed. The results are quite encouraging; they showed that our model is able to help users when searching for items.

The performance of clustering is a crucial challenge, especially for pattern recognition. The models aggregation has a positive impact on the efficiency of Data clustering. This technique is used to obtain more cluttered decision boundaries by aggregating the resulting clustering models.

Security is an integral part of the modern software systems and applications in which a client program can access different functional aspects (views) of the same domain. These views (View-Oriented Programming-VOP), as a separation of concerns approach, enable us to manage the complexity of the software systems and to accomplish greater reuse and maintainability. In VOP, an object's response to a message depends on the views currently attached to its core instance. View-oriented programming suffers from a formal model and security issues to protect the privileges of each client who needs to access different views of the same object. This paper describes a facet-based access control model to handle security issues in VOP. Especially, it introduces algebra and formalism to describe VOP and to protect the privileges of each client program. These issues are discussed through an example.

A pattern is a collection of events/features that occur together in a transaction database. Previous studies in the field are often dedicated to the problem of frequent pattern mining where only patterns that appear frequently in the input data are mined. As a result, patterns involving events/features that appear in few data sets are not captured. In some domains, such as the detection of computer attacks, fraudulent transactions in financial institutions, those patterns, also known as rare patterns, are more interesting than frequent patterns. We propose a framework to represent different categories of interesting patterns and then instantiate it to the specific case of rare patterns. Later on, we present a generic framework to mine patterns based on the Apriori approach. In this paper we are interested by the patterns composed of a set of items, also called itemsets. Thus, we instantiate the generalized Apriori framework to mine rare itemsets. The resulting approach is Apriori-like and the mine idea behind it is that if the itemset lattice representing the itemset space in classical Apriori approaches is traversed on a bottom-up manner, equivalent properties to the Apriori exploration of frequent itemsets are provided to mine rare itemsets. This include an anti-monotone property and a level- wise exploration of the itemset space. As demonstrated by our experiments, our approach is effective in identifying all rare itemsets and is more efficient than the existing approach.

Whereas the early frequent pattern mining methods admitted only relatively simple data and pattern formats (e.g., sets, sequences, etc.), there is nowadays a clear push towards the integration of ever larger portions of domain knowledge in the mining process in order to increase the precision and the abstraction level of the retrieved patterns and hence ease their interpretation. We present here a practically motivated study of a frequent pattern extraction from sequences of data objects that are described within a domain ontology. As the complexity of the descriptive structures is high, an entire framework for the pattern extraction process had to be defined. The key elements thereof are a pair of descriptive languages, one for individual data and another one for generic patterns, a generality relation between patterns, and an Apriori-like method for pattern mining.