Artificial Intelligence Review最新文献

{"title":"Towards few-call model stealing via active self-paced knowledge distillation and diffusion-based image generation","authors":"Vlad Hondru,&nbsp;Radu Tudor Ionescu","doi":"10.1007/s10462-025-11184-z","DOIUrl":"10.1007/s10462-025-11184-z","url":null,"abstract":"<div><p>Diffusion models showcase strong capabilities in image synthesis, being used in many computer vision tasks with great success. To this end, we propose to explore a new use case, namely to copy black-box classification models without having access to the original training data, the architecture, and the weights of the model, i.e. the model is only exposed through an inference API. More specifically, we can only observe the (soft or hard) labels for some image samples passed as input to the model. Furthermore, we consider an additional constraint limiting the number of model calls, mostly focusing our research on few-call model stealing. In order to solve the model extraction task given the applied restrictions, we propose the following framework. As training data, we create a synthetic data set (called proxy data set) by leveraging the ability of diffusion models to generate realistic and diverse images. Given a maximum number of allowed API calls, we pass the respective number of samples through the black-box model to collect labels. Finally, we distill the knowledge of the black-box teacher (attacked model) into a student model (copy of the attacked model), harnessing both labeled and unlabeled data generated by the diffusion model. We employ a novel active self-paced learning framework to make the most of the proxy data during distillation. Our empirical results on three data sets confirm the superiority of our framework over four state-of-the-art methods in the few-call model extraction scenario. We release our code for free non-commercial use at https://github.com/vladhondru25/model-stealing.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11184-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144135397","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":"Computational imaging based on single-photon detection: a survey","authors":"Yanyun Pu,&nbsp;Chengyuan Zhu,&nbsp;Gongxin Yao,&nbsp;Chao Li,&nbsp;Yu Pan,&nbsp;Kaixiang Yang,&nbsp;Qinmin Yang","doi":"10.1007/s10462-025-11252-4","DOIUrl":"10.1007/s10462-025-11252-4","url":null,"abstract":"<div><p>The rapid advancements in single-photon detectors with picosecond timing resolution over the past decade have significantly driven the development of time-correlated single-photon counting (TCSPC) for computational imaging applications, including bioimaging and remote sensing. In this review, we utilize the CiteSpace tool to create knowledge maps and perform a bibliometric analysis of this research area. Furthermore, we provide a comprehensive overview of the key challenges associated with computational imaging using temporal single-photon counting. We also highlight how these challenges have been addressed under extreme conditions to establish a reference model for future imaging solutions. We examine the performance evaluation parameters of single-photon detectors to enhance the understanding of detector array scaling and their application in constructing efficient computational imaging systems. Lastly, we aim to elucidate the current technical challenges in single-photon detector-based computational imaging and explore their potential future developments.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11252-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144117643","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":"Non-parametric insights in option pricing: a systematic review of theory, implementation and future directions","authors":"Akanksha Sharma,&nbsp;Chandan Kumar Verma","doi":"10.1007/s10462-025-11249-z","DOIUrl":"10.1007/s10462-025-11249-z","url":null,"abstract":"<div><p>The task of pricing options is seen as significant and receives considerable attention due to its potential to generate attractive profits through informed decision-making. Over the past few decades, researchers have extensively investigated both classical and machine-learning techniques for this purpose. Our motivation for undertaking this survey is to provide a comprehensive review and analyze systematically the recent works focusing on non-parametric models for option pricing. The analysis of the articles involves the utilization of several components such as input, output, dataset, assessment metrics, and other relevant factors. Research gaps and challenges are meticulously identified and outlined to serve as guiding insights for future improvements and advancements in the field. We categorize the implementation to assist interested researchers in easily reproducing previous studies as baselines. Based on the findings of this study, it can be inferred that the process of pricing options is a highly complicated task, requiring the consideration of several elements to enhance the accuracy and efficiency of models.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11249-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144117642","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 multi-strategy enhanced dung beetle algorithm for solving real-world engineering problems","authors":"Zhengxing Mao,&nbsp;Zhen Yang,&nbsp;Dan Luo,&nbsp;Dong Lin,&nbsp;Qinghong Jiang,&nbsp;Guoxian Huang,&nbsp;Zhixian Liao","doi":"10.1007/s10462-025-11235-5","DOIUrl":"10.1007/s10462-025-11235-5","url":null,"abstract":"<div><p>The Dung Beetle optimization (DBO) algorithm is an innovative and effective metaheuristic algorithm widely recognised for its excellent numerical optimization performance. However, DBO converges slowly and tends to fall into local optima due to the imbalance between exploration and exploitation, the lack of collaborative search capability and population diversity. To overcome these challenges, this paper proposes a dung beetle optimization algorithm based on multi-strategy collaborative enhancement (MDBO). The algorithm constructs a “search-enhance-escape” collaborative optimization framework through adaptive regulation and elite information sharing, and contains three major innovations: (1) a dual adaptive search strategy, which combines the adaptive contraction mechanism of the leader’s centre of mass and the brownian bidirectional crossover perturbation strength regulation, to enhance the population diversity and collaborative search ability of the dung beetle, achieving a dynamic balance between exploration and exploitation; (2) Elite Enhanced Solution Quality (EESQ) mechanism, which improves the quality of both the current local and global optimal positions and accelerates convergence through structured elite information and dual-phase adaptive perturbation; and (3) Dynamic Oppositional Learning (DOL), which introduces an asymmetric adaptive perturbation in the dung beetle’s foraging and Breeding phases and enhances the ability to escape from local optima. The three act synergistically to achieve a more efficient optimised search. The performance of MDBO is evaluated using the IEEE CEC 2017, CEC 2019 and CEC 2020 benchmarking functions. Compared to the DBO algorithm, the MDBO algorithm improves the convergence accuracy and stability on the CEC2017 benchmark functions by 60.91 % and 63.98 %, respectively. For the CEC2019 benchmark functions, the corresponding improvements are 54.47 % and 41.36 %, while for the CEC2020 benchmark functions, they are 50.71 % and 55.16 %, respectively. In addition, its overall performance is evaluated against two complex real-world engineering problems: UAV path planning and wireless sensor network coverage optimization. The experimental results show that MDBO provides very competitive optimization results compared to DBO, two highly referenced algorithms and five advanced algorithms.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11235-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144117638","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 LLMs and their applications in the architecture, engineering and construction industry","authors":"Dimitrios Kampelopoulos,&nbsp;Athina Tsanousa,&nbsp;Stefanos Vrochidis,&nbsp;Ioannis Kompatsiaris","doi":"10.1007/s10462-025-11241-7","DOIUrl":"10.1007/s10462-025-11241-7","url":null,"abstract":"<div><p>During the past decade, there has been rapid emergence, continuous development and advancements in the field of Artificial Intelligence (AI), and a broad adaptation ofLarge Language Models (LLMs) in a wide variety of application domains transforming and streamlining industry practices. However, the construction industry has yet to fully incorporate these technologies, delaying their wide-scale adaptation. Only a limited number of recent studies have explored the opportunities, capabilities and potential of current LLM implementations in the broad domain of Architecture Engineering and Construction (AEC) industry, leaving a significant gap in this field of research. This study aims to address this gap and provide an extensive review of already established state-of-the-art applications and use case scenarios of LLMs in the AEC industry. Apart from that, by exploring the key contributions and limitations of these applications, and by considering relative reviews on this subject, it was possible to categorize them, to extract the emerging challenges and future directions of the field and propose actionable recommendations for industry stakeholders. This study also includes an introduction to important concepts and recent advancements of LLM technologies, focusing on transformer-based architectures and providing an extensive list of LLM families.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11241-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144074011","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":"Building intelligence identification system via large language model watermarking: a survey and beyond","authors":"Xuhong Wang,&nbsp;Haoyu Jiang,&nbsp;Yi Yu,&nbsp;Jingru Yu,&nbsp;Yilun Lin,&nbsp;Ping Yi,&nbsp;Yingchun Wang,&nbsp;Yu Qiao,&nbsp;Li Li,&nbsp;Fei-Yue Wang","doi":"10.1007/s10462-025-11222-w","DOIUrl":"10.1007/s10462-025-11222-w","url":null,"abstract":"<div><p>Large Language Models (LLMs) are increasingly integrated into diverse industries, posing substantial security risks due to unauthorized replication and misuse. To mitigate these concerns, robust identification mechanisms are widely acknowledged as an effective strategy. Identification systems for LLMs now rely heavily on watermarking technology to manage and protect intellectual property and ensure data security. However, previous studies have primarily concentrated on the basic principles of algorithms and lacked a comprehensive analysis of watermarking theory and practice from the perspective of intelligent identification. To bridge this gap, firstly, we explore how a robust identity recognition system can be effectively implemented and managed within LLMs by various participants using watermarking technology. Secondly, we propose a mathematical framework based on mutual information theory, which systematizes the identification process to achieve more precise and customized watermarking. Additionally, we present a comprehensive evaluation of performance metrics for LLM watermarking, reflecting participant preferences and advancing discussions on its identification applications. Lastly, we outline the existing challenges in current watermarking technologies and theoretical frameworks, and provide directional guidance to address these challenges. Our systematic classification and detailed exposition aim to enhance the comparison and evaluation of various methods, fostering further research and development toward a transparent, secure, and equitable LLM ecosystem.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11222-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143949603","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":"Bio-inspired optimization methods for visible light communication: a comprehensive review","authors":"Martin Dratnal,&nbsp;Lukas Danys,&nbsp;Radek Martinek","doi":"10.1007/s10462-025-11251-5","DOIUrl":"10.1007/s10462-025-11251-5","url":null,"abstract":"<div><p>Visible light communication (VLC) offers a promising alternative to traditional radio frequency communication due to its greater bandwidth, energy efficiency, and security advantages. This paper presents a comprehensive review of bio-inspired optimization algorithms, including swarm intelligence and genetic algorithms, that enhance the performance and robustness of VLC systems. These techniques have demonstrated significant potential in addressing challenges such as channel optimization and noise reduction. However, despite their advantages, bio-inspired algorithms also face limitations, including computational complexity and limited adaptability to dynamic real-world conditions. Additionally, the integration of bio-inspired methods with artificial intelligence (AI) may further enhance their adaptability and efficiency in VLC systems. This review highlights both the opportunities and challenges associated with bio-inspired optimization in VLC and provides insights into future directions for research and practical implementation, which will focus on developing more efficient and scalable bio-inspired approaches that can operate in highly variable environments while minimizing energy consumption.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11251-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944280","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":"Carrier platform-enhanced multiple-UAV cooperative task assignment with dual heterogeneities","authors":"Yu Xinyong,&nbsp;Li Xin,&nbsp;Wang Lei,&nbsp;Jin Junhong,&nbsp;Zhang Genlai,&nbsp;Su Xichao,&nbsp;Tao Laifa,&nbsp;Lu Chen,&nbsp;Wang Xinwei","doi":"10.1007/s10462-025-11254-2","DOIUrl":"10.1007/s10462-025-11254-2","url":null,"abstract":"<div><p>Heterogeneous unmanned aerial vehicle (UAV) cooperation has been widely used in modern warfare. Due to the limited UAV flight endurance, the operational range is generally constrained. This issue can be effectively addressed by utilizing various airborne or shipborne carrier platforms (CPs) such as large transporters and aircraft carriers. However, such a topic is rarely studied in existing research. This paper studies the carrier platform-enhanced multiple-UAV cooperative task assignment (CPMCTA) with dual heterogeneities (i.e., in both UAVs and CPs). Additionally, the approaching unattacked target-induced risk (AUTIR), which isneglected in traditional research, is also considered to improve the task implementation safety. A novel CPMCTA model with comprehensive factors (i.e., priority, obstacles, AUTIR and heterogeneities) is first established. Aiming at an efficient solution, an adaptive self-motivated teaching-learning-based optimization algorithm (AMTLBO) is then developed by integrating various mechanisms (i.e., multiple teachers, adaptive learning rate and self-motivation). Simulations under various scenarios demonstrate the advantages of the AMTLBO in optimum-seeking capability over the other six state-of-the-art algorithms. Moreover, the necessity of considering AUTIR is highlighted. A simulation animation is available at bilibili.com/video/BV1Ht421A7Qx to provide a clearer illustration.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11254-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944281","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":"Abstracting general syntax for XAI after decomposing explanation sub-components","authors":"Stephen Wormald,&nbsp;Matheus Kunzler Maldaner,&nbsp;Kristian D. O’Connor,&nbsp;Olivia P. Dizon-Paradis,&nbsp;Damon L. Woodard","doi":"10.1007/s10462-025-11216-8","DOIUrl":"10.1007/s10462-025-11216-8","url":null,"abstract":"<div><p>Healthcare providers, policymakers, and defense contractors need to understand many types of machine learning model behaviors. While eXplainable Artificial Intelligence (XAI) provides tools for interpreting these behaviors, few frameworks, surveys, and taxonomies produce succinct yet general notation to help researchers and practitioners describe their explainability needs and quantify whether these needs are met. Such quantified comparisons could help individuals rank XAI methods by their relevance to use-cases, select explanations best suited for individual users, and evaluate what explanations are most useful for describing model behaviors. This paper collects, decomposes, and abstracts subcomponents of common XAI methods to identify a <i>mathematically grounded</i> syntax that <i>applies generally</i> to describing <i>modern and future</i> explanation types while remaining <i>useful for discovering novel XAI methods</i>. The resulting syntax, introduced as the <i>Qi</i>-Framework, generally defines explanation types in terms of the information being explained, their utility to inspectors, and the methods and information used to produce explanations. Just as programming languages define syntax to structure, simplify, and standardize software development, so too the <i>Qi</i>-Framework acts as a common language to help researchers and practitioners select, compare, and discover XAI methods. Derivative works may extend and implement the <i>Qi</i>-Framework to develop a more rigorous science for interpretable machine learning and inspire collaborative competition across XAI research.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11216-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944323","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":"High-frequency-based multi-spectral attention for domain generalization","authors":"Surong Ying,&nbsp;Xinghao Song,&nbsp;Hongpeng Wang","doi":"10.1007/s10462-025-11217-7","DOIUrl":"10.1007/s10462-025-11217-7","url":null,"abstract":"<div><p>Deep learning models have made great progress in many vision tasks, but they suffer from domain shift problem when exposed to out-of-distribution scenarios. Domain generalization (DG) is proposed to learn a model from several observable source domains that can generalize well to unknown target domains. Although recent advances in DG works have achieved promising performance, there is a high demand for computational resource, especially those that employ meta-learning or ensemble learning strategies. However, some pioneering works propose to replace convolutional neural network (CNN) as the backbone architecture with multi-layer perceptron (MLP)-like models that can not only learn long-range spatial dependencies but also reduce network parameters using Fourier transform-based techniques. Inspired by this, in this paper, we propose a high-frequency-based multi-spectral attention (HMCA) to facilitate a lightweight MLP-like model to learn global domain-invariant features by focusing on high-frequency components sufficiently. Moreover, we adopt a data augmentation strategy based on Fourier transform to simulate domain shift, thus enabling the model to pay more attention on robust features. Extensive experiments on benchmark datasets demonstrate that our method is superior to the existing CNN-based and MLP-based DG methods.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"58 8","pages":""},"PeriodicalIF":10.7,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-025-11217-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143944324","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}