Artificial Intelligence Review最新文献

Addressing the critical issue of tire wear is essential for enhancing vehicle safety, performance, and maintenance. Worn-out tires often lead to accidents, underscoring the need for effective monitoring systems. This study is vital for several reasons: safety, as worn tires increase the risk of accidents due to reduced traction and longer braking distances; performance, as uneven tire wear affects vehicle handling and fuel efficiency; maintenance costs, as early detection can prevent more severe damage to suspension and alignment systems; and regulatory compliance, as ensuring tire integrity helps meet safety regulations imposed by transportation authorities. In response, this study systematically evaluates tire conditions at 25%, 50%, 75%, and 100% wear, with an intact tire as a reference, using vibration signals as the primary data source. The analysis employs statistical, histogram, and autoregressive–moving-average (ARMA) feature extraction techniques, followed by feature selection to identify key parameters influencing tire wear. CatBoost is used for feature classification, leveraging its adaptability and efficiency in distinguishing varying wear patterns. Additionally, the study incorporates feature fusion to combine different types of features for a more comprehensive analysis. The proposed methodology not only offers a robust framework for accurately classifying tire wear levels but also holds significant potential for real-time implementation, contributing to proactive maintenance practices, prolonged tire lifespan, and overall vehicular safety.

The Rashômon Effect, applied in Explainable Machine Learning, refers to the disagreement between the explanations provided by various attribution explainers and to the dissimilarity across multiple explanations generated by a particular explainer for a single instance from the dataset (differences between feature importances and their associated signs and ranks), an undesirable outcome especially in sensitive domains such as healthcare or finance. We propose a method inspired from textual-case based reasoning for aligning explanations from various explainers in order to resolve the disagreement and dissimilarity problems. We iteratively generated a number of 100 explanations for each instance from six popular datasets, using three prevalent feature attribution explainers: LIME, Anchors and SHAP (with the variations Tree SHAP and Kernel SHAP) and consequently applied a global cluster-based aggregation strategy that quantifies alignment and reveals similarities and associations between explanations. We evaluated our method by weighting the (:k)-NN algorithm with agreed feature overlap explanation weights and compared it to a non-weighted (:k)-NN predictor, having as task binary classification. Also, we compared the results of the weighted (:k)-NN algorithm using aggregated feature overlap explanation weights to the weighted (:k)-NN algorithm using weights produced by a single explanation method (either LIME, SHAP or Anchors). Our global alignment method benefited the most from a hybridization with feature importance scores (information gain), that was essential for acquiring a more accurate estimate of disagreement, for enabling explainers to reach a consensus across multiple explanations and for supporting effective model learning through improved classification performance.

Controllable Image Synthesis (CIS) is a methodology that allows users to generate desired images or manipulate specific attributes of images by providing precise input conditions or modifying latent representations. In recent years, CIS has attracted considerable attention in the field of image processing, with significant advances in consistency, controllability and harmony. However, several challenges still remain, particularly regarding the fine-grained controllability and interpretability of synthesized images. In this paper, we comprehensively and systematically review the CIS from problem definition, taxonomy and evaluation systems to existing challenges and future research directions. First, the definition of CIS is given, and several representative deep generative models are introduced in detail. Second, the existing CIS methods are divided into three categories according to the different control manners used and discuss the typical work in each category critically. Furthermore, we introduce the public datasets and evaluation metrics commonly used in image synthesis and analyze the representative CIS methods. Finally, we present several open issues and discuss the future research direction of CIS.

Atmospheric boundary layer (ABL) structure and dynamics are important aspects to consider in human health. The ABL is characterized by a high degree of spatial and temporal variability that hinders their understanding. This paper aims to provide a comprehensive overview of machine learning (ML) methodologies, encompassing deep learning and ensemble approaches, within the scope of ABL research. The goal is to highlight the challenges and opportunities of using ML in turbulence modeling and parameterization in areas such as atmospheric pollution, meteorology, and renewable energy. The review emphasizes the validation of results to ensure their reliability and applicability. ML has proven to be a valuable tool for understanding and predicting how ABL spatial and seasonal variability affects pollutant dispersion and public health. In addition, it has been demonstrated that ML can be used to estimate several variables and parameters, such as ABL height, making it a promising approach to enhance air quality management and urban planning.

One of the focal points in industrial product defect detection lies in the utilization of deep learning-based object detection algorithms. With the continuous introduction of these algorithms and their refined models, notable achievements have been attained. However, challenges persist in industrial settings, such as substantial variations in defect scales, the delicate balance between accuracy and speed, and the detection of small objects. Various methods have been proposed to address these challenges and propel the advancement of defect detection. To comprehensively review the latest developments in deep learning-based industrial product defect detection algorithms and foster further progress, this paper encompasses typical datasets and evaluation metrics used in industrial product defect detection, traces the development history of supervised one-stage and two-stage object detection algorithm-based and unsupervised algorithm-based industrial defect detection methods, discusses major challenges, and outlines future directions. It highlights the potential for further improving the accuracy, speed, and reliability of defect detection systems in industrial applications.

The African Vultures Optimization Algorithm (AVOA) is a recently developed meta-heuristic algorithm inspired by the foraging behavior of African vultures in nature. This algorithm has gained attention due to its simplicity, flexibility, and effectiveness in tackling many optimization problems. The significance of this review lies in its comprehensive examination of the AVOA’s development, core principles, and applications. By analyzing 112 studies, this review highlights the algorithm’s versatility and the growing interest in enhancing its performance for real-world optimization challenges. This review methodically explores the evolution of AVOA, investigating proposed improvements that enhance the algorithm’s ability to adapt to various search geometries in optimization problems. Additionally, it introduces the AVOA solver, detailing its functionality and application in different optimization scenarios. The review demonstrates the AVOA’s effectiveness, particularly its unique weighting mechanism, which mimics vulture behavior during the search process. The findings underscore the algorithm’s robustness, ease of use, and lack of dependence on derivative information. The review also critically evaluates the AVOA’s convergence behavior, identifying its strengths and limitations. In conclusion, the study not only consolidates the existing knowledge on AVOA but also proposes directions for future research, including potential adaptations and enhancements to address its limitations. The insights gained from this review offer valuable guidance for researchers and practitioners seeking to apply or improve the AVOA in various optimization tasks.

Abductive logic programming (ALP) extends logic programming with hypothetical reasoning by means of abducibles, an extension able to handle interesting problems, such as diagnosis, planning, and verification with formal methods. Implementations of this extension have been using Prolog meta-interpreters and Prolog programs with Constraint Handling Rules (CHR). While the latter adds a clean and efficient interface to the host system, it still suffers in performance for large programs. Here, the concern is to obtain a more performant implementation of the SCIFF system following a compiled approach. This paper, as a first step in this long term goal, sets out a propositional ALP system following SCIFF, eliminating the need for CHR and achieving better performance.

In response to the shortcomings of particle swarm optimization (PSO), such as low execution efficiency and difficulty in overcoming local optima, this paper proposes a multi-strategy PSO method incorporating snow ablation operation (SAO), known as SAO-MPSO. Firstly, Cubic initialization is performed on particles to obtain a good initial environment. Subsequently, SAO and PSO are combined in parallel, and a balanced search mechanism led by multiple sub-populations is devised, significantly improving the search efficiency of overall population. Finally, the degree day method of SAO is introduced, and particles are endowed with memory of environmental changes to prevent premature convergence of PSO, while balancing the exploration and exploitation (ENE) capabilities in later phases. All adaptive parameters are used throughout this method in place of fixed parameters to improve the robustness and adaptability. For a comprehensive analysis of SAO-MPSO, its good ENE ability is verified on CEC 2020 and CEC 2022 and this method is compared with existing improved PSO versions on both test sets. The results show that SAO-MPSO has certain advantages in the comparison of similar improved algorithms. In order to further validate the strength of SAO-MPSO in dealing with nonlinear optimization problems (OPs) with strong constraints, firstly, based on the ball Wang-Ball (BWB) curve, a combined BWB (CBWB) curve is constructed, and a construction method for CBWB curves that satisfy G¹ and G² continuity is derived. Then, with the energy minimization and scale parameters of the CBWB curve as the optimization objective and variables respectively, a shape optimization model that satisfies G² continuity is established. Finally, three numerical optimization examples based on this model are solved using SAO-MPSO and compared with 10 other methods. The results show that the energy obtained by SAO-MPSO is the smallest, which verifies the effectiveness of this method applied to shape OPs of CBWB curve.

In the rapidly evolving domain of cybersecurity, the imperative for intrusion detection systems is undeniable; yet, it is increasingly clear that to meet the ever-growing challenges posed by sophisticated threats, intrusion detection itself stands in need of the transformative capabilities offered by the explainable artificial intelligence (xAI). As this concept is still developing, it poses an array of challenges that need addressing. This paper discusses 25 of such challenges of varying research interest, encountered in the domain of xAI, identified in the course of a targeted study. While these challenges may appear as obstacles, they concurrently present as significant research opportunities. These analysed challenges encompass a wide spectrum of concerns spanning the intersection of xAI and cybersecurity. The paper underscores the critical role of xAI in addressing opacity issues within machine learning algorithms and sets the stage for further research and innovation in the quest for transparent and interpretable artificial intelligence that humans are able to trust. In addition to this, by reframing these challenges as opportunities, this study seeks to inspire and guide researchers towards realizing the full potential of xAI in cybersecurity.

Alzheimer’s disease (AD) is a growing global concern, exacerbated by an aging population and the high costs associated with traditional detection methods. Recent research has identified speech data as valuable clinical information for AD detection, given its association with the progressive degeneration of brain cells and subsequent impacts on memory, cognition, and language abilities. The ongoing demographic shift toward an aging global population underscores the critical need for affordable and easily available methods for early AD detection and intervention. To address this major challenge, substantial research has recently focused on investigating speech data, aiming to develop efficient and affordable diagnostic tools that align with the demands of our aging society. This paper presents an in-depth review of studies from 2018–2023 utilizing speech for AD detection. Following the PRISMA protocol and a two-stage selection process, we identified 85 publications for analysis. In contrast to previous literature reviews, this paper places a strong emphasis on conducting a rigorous comparative analysis of various Artificial Intelligence (AI) based techniques, categorizing them meticulously based on underlying algorithms. We perform an exhaustive evaluation of research papers leveraging common benchmark datasets, specifically ADReSS and ADReSSo, to assess their performance. In contrast to previous literature reviews, this work makes a significant contribution by overcoming the limitations posed by the absence of standardized tasks and commonly accepted benchmark datasets for comparing different studies. The analysis reveals the dominance of deep learning models, particularly those leveraging pre-trained models like BERT, in AD detection. The integration of acoustic and linguistic features often achieves accuracies above 85%. Despite these advancements, challenges persist in data scarcity, standardization, privacy, and model interpretability. Future directions include improving multilingual recognition, exploring emerging multimodal approaches, and enhancing ASR systems for AD patients. By identifying these key challenges and suggesting future research directions, our review serves as a valuable resource for advancing AD detection techniques and their practical implementation.