Artificial Intelligence Review最新文献

{"title":"Novel coefficients for improved robustness in multi-criteria decision analysis","authors":"Bartosz Paradowski,&nbsp;Jarosław Wątróbski,&nbsp;Wojciech Sałabun","doi":"10.1007/s10462-025-11307-6","DOIUrl":"10.1007/s10462-025-11307-6","url":null,"abstract":"<div><p>In multi-criteria decision-making (MCDM), decision-makers face increasing complexity and the need for enhanced tools to facilitate informed and well-aligned decision outcomes. A critical challenge in MCDM is the determination of criteria weights, which significantly influence the final ranking of alternatives. While recent approaches aim to eliminate the need for explicit weight assignment, certain decision contexts necessitate their inclusion. This study introduces two novel coefficients, Rank Stability (RS) and Balance Point (BP), designed to provide deeper insights into the decision problem and its solution properties. Rank Stability quantifies the robustness of a solution against perturbations, while Balance Point evaluates the conditioning of the solution within the problem’s structure. The decision problem is defined by a set of alternatives and criteria, where modifications to alternatives require a reassessment of the decision model. To examine the properties of these coefficients, this study employs simulation experiments utilizing the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and VIseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) methods, alongside case-based analyses demonstrating their practical applications. Additionally, extreme cases of RS and BP values are explored to enhance interpretability for decision-makers. A real-world decision problem is further analyzed to illustrate the applicability of these coefficients and introduce a novel framework for comparing MCDM methodologies. This approach facilitates a more systematic and comprehensive assessment of MCDM methods, contributing to the advancement of decision-support tools.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11307-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145162248","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neural architecture search: two constant shared weights initialisations","authors":"Ekaterina Gracheva","doi":"10.1007/s10462-025-11238-2","DOIUrl":"10.1007/s10462-025-11238-2","url":null,"abstract":"<div><p>In the last decade, zero-cost metrics have gained prominence in neural architecture search (NAS) due to their ability to evaluate architectures without training. These metrics are significantly faster and less computationally expensive than traditional NAS methods and provide insights into neural architectures’ internal workings. This paper introduces <span>epsinas</span>, a novel zero-cost NAS metric that assesses architecture potential using two constant shared weight initialisations and the statistics of their outputs. We show that the dispersion of raw outputs, normalised by their average magnitude, strongly correlates with trained accuracy. This effect holds across image classification and language tasks on NAS-Bench-101, NAS-Bench-201, and NAS-Bench-NLP. Our method requires no data labels, operates on a single minibatch, and eliminates the need for gradient computation, making it independent of training hyperparameters, loss metrics, and human annotations. It evaluates a network in a fraction of a GPU second and integrates seamlessly into existing NAS frameworks. The code supporting this study can be found on GitHub at https://github.com/egracheva/epsinas.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11238-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145162242","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Techniques and challenges for nuclei segmentation in cervical smear images: a review","authors":"Assad Rasheed,&nbsp;Syed Hamad Shirazi,&nbsp;Pordil Khan,&nbsp;Ali M. Aseere,&nbsp;Muhammad Shahzad","doi":"10.1007/s10462-025-11207-9","DOIUrl":"10.1007/s10462-025-11207-9","url":null,"abstract":"<div><p>Cervical cancer is one of the fastest-growing cancers affecting women, leading to a significant number of deaths. However, early detection and timely treatment can greatly reduce the mortality rate and improve the chances of recovery. A widely used method for the diagnosis of cancer is the manual analysis of tissue biopsy specimens on slides. This manual process of specimen examination is time-consuming and error-prone, resulting in an increasing interest in digitizing histopathological workflows to refine and expedite analysis. This paper has compiled a comprehensive review of automated cervical nuclei segmentation approaches in histopathological images. We have examined both deep learning and traditional image segmentation methods, working mechanisms, and their variants, as well as the datasets used to evaluate these approaches. To find relevant studies, we searched on platforms such as IEEE Xplore, Google Scholar, ACM Digital Library, SpringerLink, and ScienceDirect using keywords such as cervical nuclei, nuclear, and nucleus, deep learning network (DNN), convolutional neural networks (CNNs), cervical cytology or histopathology, and traditional or classical image segmentation methods. We reviewed 78 research papers on both classical image segmentation and deep learning-based techniques specifically designed for nuclei segmentation in cervical histopathological images, published from 2010 until October 2024. We organized these studies into two main categories: Classical image segmentation methods and deep learning approaches, subdividing them into relevant subcategories and compiled a comparative analysis of their results, identified ongoing challenges in nuclei segmentation, and highlighted opportunities and future prospects for this task. We discussed recent studies on cervical nuclei segmentation using automated methods. The implementation of automated image segmentation methods and their various extensions has significantly improved the performance of automated diagnostic systems for cervical cancer.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11207-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145161417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A review of feature selection methods for actual evapotranspiration prediction","authors":"Chalachew Muluken Liyew,&nbsp;Stefano Ferraris,&nbsp;Elvira Di Nardo,&nbsp;Rosa Meo","doi":"10.1007/s10462-025-11298-4","DOIUrl":"10.1007/s10462-025-11298-4","url":null,"abstract":"<div><p>Accurate prediction of actual evapotranspiration (AET) is critical for hydrological modeling, agricultural planning, and climate studies. Machine learning models have emerged as powerful AET prediction tools because they can handle complex, nonlinear relationships in large datasets. However, selecting relevant input features significantly impacts model performance, efficiency, and interpretability. Feature selection techniques reduce high-dimensional datasets by identifying redundant and uncorrelated variables. This paper reviews feature selection approaches for predicting ML-based AETs by analyzing 62 studies; a total of 416 were retrieved from seven digital libraries. Our analysis shows that filtering methods are the most widely used <span>((38.8%))</span>, followed by manual selection based on domain expertise <span>((28.7%))</span>, embedded methods <span>((17.5%))</span>, and wrapper methods <span>((11.2%))</span>. Dimensionality reduction techniques, such as principal component analysis (PCA), are the least used <span>((3.8%))</span>. Among machine learning models, Random Forest (RF) and Artificial Neural Networks (ANN) are the most commonly used, with 29 and 27 instances, respectively. The study highlights the strengths and limitations of each category of feature selection, emphasizing the potential of hybrid approaches integrating filter, wrapper, embedded, and manual selection methods. These combinations improve model accuracy, robustness, and generalization, while mitigating overfitting, computational inefficiency, and sensitivity to noise. This review provides insights into optimal feature selection strategies for improving ML-based AET prediction.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11298-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145161423","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Peeping at creAItivity through a keyhole: creative self-perceptions, potential, and enhancement of GenAI chatbots","authors":"Dimitris Grammenos,&nbsp;Todd Lubart","doi":"10.1007/s10462-025-11288-6","DOIUrl":"10.1007/s10462-025-11288-6","url":null,"abstract":"<div><p>The work in this paper investigates (a) the (emerging) creative self-perceptions of GenAI chatbots, (b) their creative potential, (c) their ability to self-assess the creativity of their own outcomes and that of their peers, and (d) how their creative outcomes can be improved. To this end, an exploratory study was implemented involving three popular commercial chatbots: ChatGPT (GPT-4o - paid) by OpenAI, Claude (3.5 Sonnet - paid) by Anthropic and Gemini (1.5 Flash - free) by Google. The study included four phases and employed well-established methods and tools from the scientific domain of (human) creativity research, including the Short Scale of Creative Self (SSCS) questionnaire, a verbal test of convergent-integrative thought from the Evaluation of Potential Creativity (EPoC) battery which was scored by the chatbots and human experts, and the Dynamic Assessment (DA) approach and humor as means for enhancing the chatbots’ creative outcomes. The results of the study are encapsulated in 21 original and, sometimes, surprising observations and in 6 practical insights regarding the use of GenAI chatbots as creativity-support tools.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11288-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145161422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Script identification in multilingual environment: a survey in recent years","authors":"Yaowei Yang,&nbsp;Elham Eli,&nbsp;Alimjan Aysa,&nbsp;Kurban Ubul","doi":"10.1007/s10462-025-11194-x","DOIUrl":"10.1007/s10462-025-11194-x","url":null,"abstract":"<div><p>Multilingualism is an important trend in the field of optical character recognition (OCR). In a multilingual environment, the task of script identification often combines with other tasks to complete multilingual work jointly. As a front-end function of a multilingual OCR system, it automatically identifies the language of the text image and further recognizes text in multilingual engines. In reality, script identification plays a major role, especially in multilingual scene understanding, as well as intelligent document analysis and recognition. This survey introduces the technology of script identification and summarizes the related work developed in this field from 2017 to date, including traditional learning, deep learning, and available datasets. Based on a comprehensive analysis of existing work, it provides a new survey for researchers to grasp recent script identification work. By discussing the problems that need to be solved, it can lay the foundation for related research work and activities.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 10","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11194-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145162167","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Image enhancement framework combining interval-valued intuitionistic fuzzy sets and fractional sobel operator","authors":"Ravindar Raj Chinnappan,&nbsp;Dhanasekar Sundaram","doi":"10.1007/s10462-025-11294-8","DOIUrl":"10.1007/s10462-025-11294-8","url":null,"abstract":"<div><p>Image enhancement in low-light conditions is a difficult challenge due to noise, visual impairment and colour distortion. This research describes a new method for improving low-light images utilizing fuzzy-based and fractional approach. Firstly, normalize the low-light image to minimize noise and increase clarity, resulting a fuzzy image. The fuzzy image is then turned into an intuitionistic fuzzy image (IFI), which considers membership and non-membership values, presenting a more accurate characterization of uncertainty in pixel intensities. The IFI eventually transforms into an interval-valued intuitionistic fuzzy image (IVIFI) which captures a broader range of uncertainty. A fractional Sobel mask is then used to convolute the IVIF image, resulting in an improved accurate intensity distribution. The result is an optimally enhanced image with excellent contrast and detail preservation. This proposed method gives highest values of entropy, contrast improvement index, absolute mean brightness error and colourfulness are 7.9017, 6.4658, 125.5033 and 0.2853 respectively. A comparison with existing approaches indicates the proposed method’s superiority in visual quality and quantitative metrics, emphasizing its efficacy in improving low-light images.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 9","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11294-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145142222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multi-view learning with graph convolution networks adopting diverse graphs and genuine deep feature fusion","authors":"Guowen Peng,&nbsp;Fadi Dornaika,&nbsp;Jinan Charafeddine","doi":"10.1007/s10462-025-11301-y","DOIUrl":"10.1007/s10462-025-11301-y","url":null,"abstract":"<div><p>Multi-view data significantly enhances the accuracy of machine learning algorithms by providing a comprehensive representation of object features. Despite their potential, research on the use of Graph Convolutional Networks (GCNs) for processing node connectivity and data features remains limited. Existing methods mainly focus on weighted summation of graph matrices, with only a few approaches effectively integrating the feature information into the graph structures. To overcome these limitations, this paper proposes a novel deep learning architecture: the Feature Fusion and Multi-Graph Fusion Learning Framework (MGCN-FN). The framework consists of two core modules: Feature Fusion Network (FFN): Designed to extract and consolidate key features from multiple views. Multi-Graph Fusion Network (MGFN): Constructs multiple graphs for each view and jointly optimizes both the graph weights and the GCN model. Extensive experiments on various multi-view datasets show that MGCN-FN achieves superior performance compared to state-of-the-art methods, especially on semi-supervised multi-view classification tasks.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 9","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11301-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141907","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"General information metrics for improving AI model training efficiency","authors":"Jianfeng Xu,&nbsp;Congcong Liu,&nbsp;Xiaoying Tan,&nbsp;Xiaojie Zhu,&nbsp;Anpeng Wu,&nbsp;Huan Wan,&nbsp;Weijun Kong,&nbsp;Chun Li,&nbsp;Hu Xu,&nbsp;Kun Kuang,&nbsp;Fei Wu","doi":"10.1007/s10462-025-11281-z","DOIUrl":"10.1007/s10462-025-11281-z","url":null,"abstract":"<div><p>To address the growing size of AI model training data and the lack of a universal data selection methodology–factors that significantly drive up training costs–this paper presents the General Information Metrics Evaluation (GIME) method. GIME leverages general information metrics from Objective Information Theory (OIT), including <i>volume</i>, <i>delay</i>, <i>scope</i>, <i>granularity</i>, <i>variety</i>, <i>duration</i>, <i>sampling rate</i>, <i>aggregation</i>, <i>coverage</i>, <i>distortion</i>, and <i>mismatch</i> to optimize dataset selection for training purposes. Comprehensive experiments conducted across diverse domains, such as CTR Prediction, Civil Case Prediction, and Weather Forecasting, demonstrate that GIME effectively preserves model performance while substantially reducing both training time and costs. Additionally, applying GIME within the Judicial AI Program led to a remarkable 39.56% reduction in total model training expenses, underscoring its potential to support efficient and sustainable AI development.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 9","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11281-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145141908","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Estimation of photovoltaic parameters by dynamic updating and selecting a snake optimizer based on multi-directional optimization","authors":"Yi Wang,&nbsp;Fushuai Ping,&nbsp;Yuchen Li,&nbsp;Tianfeng Gu,&nbsp;Tan Wang","doi":"10.1007/s10462-025-11272-0","DOIUrl":"10.1007/s10462-025-11272-0","url":null,"abstract":"<div><p>A multi-strategy synergistic learning snake optimizer (MSSO) is proposed to address the challenges of high computational cost, limited identification of key parameters, and inaccurate results that are still prevalent in swarm intelligence algorithms when managing the nonlinear dynamics of solar photovoltaic systems. This paper, for the first time, explores the intrinsic mechanisms between the four behavioral patterns generated by snakes, which are influenced by food quantity and temperature, and their impact on the diversity and convergence of snake optimization algorithms. It also discusses the limitations observed, offering a novel interpretation of snake optimization algorithms from a fresh perspective. The innovatively proposed superior point strategy, adaptive snake spiral foraging strategy, and dynamic update selection mechanism with multi-directional optimization enable the algorithm to learn from the global optimum and the neighborhood optimum, reducing individuals’ over-reliance on optimal positions and accelerating convergence. Extensive experiments are conducted using CEC2017 and CEC2011 benchmarks on 43 function problems and three application problems for photovoltaic parameter estimation to evaluate the performance of MSSO. A comparative analysis with 26 metaheuristic algorithms (MAs) indicates that MSSO converges more rapidly and ranks first in terms of mean, standard deviation, and Wilcoxon and Friedman tests. The results of the half violin plot combined with scatter plots further illustrate that MSSO exhibits a denser data cloud, a more concentrated distribution density of optimal values, fewer outliers, and enhanced stability. Additionally, a higher quality solution can be obtained with only 50% of the iterations required, without any additional computational time. Finally, on the three application problems of photovoltaic parameter estimation, compared to 26 MAs, the solutions provided by MSSO ranked first in terms of mean and Root Mean Square Error (RMSE), and the performance of the algorithms can be improved by up to 94.0% and 2.7%, which highlights the superiority, universality, and applicability of the algorithms.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 9","pages":""},"PeriodicalIF":13.9,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11272-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145142101","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}