Artificial Intelligence Review最新文献

{"title":"Embedding learning on multiplex networks for link prediction","authors":"Orell Trautmann,&nbsp;Olaf Wolkenhauer,&nbsp;Clémence Réda","doi":"10.1007/s10462-026-11656-w","DOIUrl":"10.1007/s10462-026-11656-w","url":null,"abstract":"<div><p>Over the past years, embedding learning on networks has shown tremendous results in link prediction tasks for complex systems, with a wide range of real-life applications. Learning a representation for each node in a knowledge graph allows us to capture topological and semantic information, which can be processed in downstream analyses later. In the link prediction task, high-dimensional network information is encoded into low-dimensional vectors, which are then fed to a predictor to infer new connections between nodes in the network. As the network complexity (that is, the numbers of connections and types of interactions) grows, embedding learning turns out increasingly challenging. This review covers published models on embedding learning on multiplex networks for link prediction. First, we propose refined taxonomies to classify and compare 38 models (in 36 papers) across five methodological categories and three embedding classes. Second, we review and address the problem of reproducible and fair evaluation of embedding learning on multiplex networks for the link prediction task. Finally, we highlight challenges specific to performance evaluation on directed multiplex networks, and address them by introducing a novel and fair testing procedure. This review constitutes a crucial step towards the development of more performant and tractable embedding learning approaches for multiplex networks and their fair evaluation for the link prediction task. We also suggest guidelines on the evaluation of models, and provide an informed perspective on the challenges and tools currently available to address downstream analyses applied to multiplex networks.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 9","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11656-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148750687","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":"All too perfect: bias and aspiration in persona generation with LLMs","authors":"Nicholas Kluge Corrêa,&nbsp;Rafaela Weber Mallmann,&nbsp;David Kaczér,&nbsp;Florian Mai,&nbsp;Ana Ilievska,&nbsp;Julia Maria Mönig","doi":"10.1007/s10462-026-11641-3","DOIUrl":"10.1007/s10462-026-11641-3","url":null,"abstract":"<div><p>Synthetic data generated by large language models plays a central role in the training and alignment process of other AI systems. However, this process also risks inheriting the structural biases of organic corpora and embedding new biases that stem from the design choices underlying the data creation process. This paper examines the systematic biases that emerge when large language models (LLMs) are tasked with generating synthetic personas. We introduce a reproducible, minimally conditioned pipeline that produced 40,000 personas, in four different languages, using two instruction-tuned open-weight generators (Llama−3.3-70B-Instruct and Qwen2.5-72B-Instruct), followed by a battery of quantitative analyses: name match-rates, KL divergence/skew for gender, age-pyramid comparisons, profession-gender intersectionals, adjective/sentiment profiling, and Proppian role classification. Our main findings reveal that persona generations are far from neutral. Models tend to focus on middle-aged, aspirational, and overwhelmingly positive (i.e., upbeat/optimistic) characters, while non-binary identities and many real-world occupations remain underrepresented. We conclude that contemporary training and alignment regimes produce a form of narrative sanitization that both flattens representational diversity and embeds normative assumptions, and propose that persona-based evaluation can serve as a scalable diagnostic of what generative systems <i>value</i> and <i>prioritize</i> when depicting humanity.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 9","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11641-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148613172","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":"Evaluating large language model compression: a comparative analysis on state-of-the-art models across diverse hardware platforms","authors":"Dominik Hildebrand,&nbsp;Benjamin Kiefer,&nbsp;Andreas Zell","doi":"10.1007/s10462-026-11614-6","DOIUrl":"10.1007/s10462-026-11614-6","url":null,"abstract":"<div><p>This work presents a systematic, empirical comparison of contemporary compression techniques for large language models (LLMs), namely quantization, pruning, and parameter-efficient fine-tuning (PEFT) using a representative set of open-source model families (Llama, Mistral, Phi and Qwen) and model scales (1.7 Billion to 70 Billion). Evaluation combined benchmarks (MMLU, SQuAD v2, TinyBenchmarks and WikiText), deployment metrics (peak memory, time-to-first-token, tokens/sec and maximum sequence lengths) and settings (multi-GPU clusters, single-GPU PC, laptop, and smartphone) to capture real-world trade-offs. Quantization often delivered the best wins for deployment feasibility—enabling single-device and mobile inference—but required careful per-model tuning and backend support to avoid throughput regressions. Pruning reduced parameter counts substantially but frequently incured large, even catastrophic, performance loss beyond moderate sparsity levels. Retraining partially mitigated this but did not uniformly close the gap to quantization. Finally, PEFT methods enabled models to match or outperform models with up to 18 times the parameters on SQuAD v2 while reducing storage as well as optimizer overhead and often improved task performance even when full fine-tuning failed.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 9","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11614-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148433596","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":"Mitigating cyberattacks on autonomous vehicles: a comprehensive review of Generative Artificial Intelligence defense techniques","authors":"May Phyu Phyu Thaw,&nbsp;Doreen Sebastian Sarwatt,&nbsp;Huansheng Ning,&nbsp;Jianguo Ding","doi":"10.1007/s10462-026-11636-0","DOIUrl":"10.1007/s10462-026-11636-0","url":null,"abstract":"<div><p>Autonomous vehicles (AVs) are rapidly becoming foundational components of intelligent transportation systems (ITS), yet their complex cyber-physical architectures expose them to a broad and continuously evolving threat landscape. Existing cybersecurity solutions struggle to keep pace with the dynamic, data-intensive nature of AV ecosystems, leaving critical vulnerabilities unaddressed across perception, communication, and decision-making subsystems. Generative Artificial Intelligence (GAI), encompassing Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Diffusion Models (DMs), has emerged as a powerful paradigm for both offensive simulation and defensive reinforcement, enabling synthetic data generation, adversarial attack emulation, and enhanced anomaly and intrusion detection. Yet despite growing interest in GAI for general cybersecurity, its systematic application to AV-specific security remains fragmented and underexplored. This paper addresses that gap through a PRISMA-guided systematic review of GAI-driven defense mechanisms for AV cybersecurity, synthesizing 216 peer-reviewed studies drawn from major scientific databases and published between January 2020 and February 2026. Three principal contributions are made. First, we introduce an AV-centric, three-dimensional taxonomy that classifies defenses along generative architecture, defensive function, and AV-relevant attack surface, explicitly anchoring each study to AV subsystems and operational contexts. Second, we provide a disciplined synthesis that separates study-specific performance findings from broader design insights, exposing fundamental gaps between conventional and GAI-based approaches in scalability, adaptability, and resilience. Third, we identify critical open challenges—including training instability, the absence of standardized AV security benchmarks, real-time deployment constraints, and limited explainability—and propose targeted research directions for safety-critical environments. By grounding GAI defenses within AV system layers and cyber-physical threat models, this review serves as a practitioner- and researcher-oriented reference for building robust, scalable, and trustworthy cybersecurity solutions for next-generation autonomous vehicles.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 9","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11636-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148433902","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 on empirical studies in explainable artificial intelligence","authors":"Tobias Jahn,&nbsp;Philipp Hühn,&nbsp;Manfred Reichert","doi":"10.1007/s10462-026-11595-6","DOIUrl":"10.1007/s10462-026-11595-6","url":null,"abstract":"<div><p>As artificial intelligence (AI) systems become more integrated into decision-making processes, the need for explainability has emerged to foster trust, understanding, and effective human-AI collaboration. With the variety of explainable AI (XAI) methods available, selecting the right one for a specific user group and a given use case remains challenging, especially given the limited empirical validation of existing theoretical guidance. This systematic literature review addresses this gap by synthesizing human-grounded evaluations of XAI methods to identify the influence of specific explanation properties on user outcomes across diverse settings. Moving beyond high-level taxonomies, we classify XAI methods along multiple dimensions, such as scope and output type. Based on this classification, we analyze how the properties of XAI methods affect different user groups, tasks, and domains. Our findings underscore the necessity of context-aware method selection, as the effectiveness of XAI methods varies significantly across use cases. Moreover, our analysis reveals imbalances in the existing empirical landscape, where certain methods and user groups are overrepresented while others are largely overlooked. By validating theoretical proposals with empirical evidence, this review provides actionable guidance for selecting XAI methods that are aligned with user needs and use case demands, paving the way for more targeted and effective human-AI interaction.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 8","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11595-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148380636","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":"From controlled benchmarks to unconstrained environments: a pipeline-oriented survey of models, domains, and emerging trends on human activity recognition","authors":"Thiago Brito Cassimiro da Silva,&nbsp;Hemir da Cunha Santiago,&nbsp;Bruno José Torres Fernandes","doi":"10.1007/s10462-026-11619-1","DOIUrl":"10.1007/s10462-026-11619-1","url":null,"abstract":"<div><p>Human Activity Recognition (HAR) has become a central research topic in computer vision, machine learning, and pervasive computing, driven by its wide range of applications in intelligent surveillance, healthcare monitoring, human–computer interaction, and smart environments. Over the past two decades, the field has evolved from handcrafted feature-based approaches and classical statistical models to deep learning architectures and hybrid paradigms that combine representation learning with robust classifiers. This paper presents a comprehensive survey of HAR research from 2004 to February 2026, systematically organizing the literature according to a canonical processing pipeline that includes sensing modalities/domains, feature representations, temporal modeling strategies, learning paradigms, datasets, and evaluation protocols. Beyond a descriptive review, the survey provides a cross-analysis of the relationships between techniques, classifiers, datasets, and application domains, highlighting recurring design patterns, trade-offs, and performance implications. Particular attention is given to emerging trends such as hybrid deep–classical models, privacy-preserving HAR, device-free sensing, and edge-oriented deployments. By synthesizing methodological advances and practical considerations, this survey aims to serve both as an entry point for new researchers and as a decision-oriented reference for practitioners seeking to design robust, generalizable, and ethically responsible HAR systems.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 8","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11619-1.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148380353","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":"AI safety landscape for large language models: taxonomy, state-of-the-art, and future directions","authors":"Chen Chen,&nbsp;Xueluan Gong,&nbsp;Ziyao Liu,&nbsp;Weifeng Jiang,&nbsp;Si Qi Goh,&nbsp;Kwok-Yan Lam","doi":"10.1007/s10462-026-11590-x","DOIUrl":"10.1007/s10462-026-11590-x","url":null,"abstract":"<div><p>AI safety is an emerging field of critical importance for the secure adoption and deployment of AI systems. With the recent advancements in large language models (LLMs), the technological landscape surrounding the design, development, and deployment of AI systems has undergone significant change. The failure of AI systems at one organization, or AI risks undertaken by one organization, can propagate down the AI technology supply chain, affect the entire AI ecosystem, and potentially lead to collective failures and cause large-scale harm to society. In this paper, we propose a novel architectural framework for understanding and analyzing AI safety in the context of LLMs, defining its characteristics through three key perspectives: Trustworthy AI, Responsible AI, and Ecosystemic Safe AI. We provide a comprehensive review of current research and advancements in AI safety from these perspectives, identifying major challenges and outlining mitigation strategies. Additionally, we highlight potential future directions that warrant further exploration to advance AI safety research and, ultimately, strengthen public trust in digital transformation.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 10","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11590-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148837757","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":"Advancing integrated sensing and communications with RL/DRL in 6 G networks: a survey","authors":"Wali Ullah Khan,&nbsp;Waqas Khalid,&nbsp;Muhammad Iqbal,&nbsp;Chiew Foong Kwong,&nbsp;Manzoor Ahmed,&nbsp;Syed Tariq Shah","doi":"10.1007/s10462-026-11610-w","DOIUrl":"10.1007/s10462-026-11610-w","url":null,"abstract":"<div><p>Integrated sensing and communication (ISAC) has emerged as a key enabling technology for sixth-generation (6 G) networks, supporting joint environment perception and data transmission with high spectral efficiency and reduced hardware cost. However, the design and deployment of ISAC systems remain challenging due to dynamic wireless environments, sensing–communication trade-offs, and the increasing complexity of large-scale networks. Reinforcement learning (RL) and deep reinforcement learning (DRL) have recently attracted significant attention as data-driven approaches for addressing these challenges through model-free, adaptive, and real-time optimization. This survey first reviews the fundamentals of ISAC technology and then summarizes major RL and DRL algorithms relevant to ISAC design. It further provides a comprehensive overview of RL/DRL methods for ISAC in 6 G networks. Existing approaches are categorized into value-based, policy-based, and hybrid methods, and are further classified according to representative ISAC scenarios, including reconfigurable intelligent surface (RIS)-assisted systems, unmanned aerial vehicle (UAV) and satellite systems, vehicular networks, and other emerging 6 G applications. The survey highlights reported gains in spectral efficiency, sensing accuracy, adaptability, and robustness, while also identifying key limitations of current approaches. Finally, the survey outlines current challenges, future research directions, and key lessons learned.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 10","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11610-w.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148782496","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":"Toward a unified governing framework for modern artificial intelligence: a taxonomy-based cross-domain survey","authors":"Abbas Amini,&nbsp;Narjes Firouzkouhi,&nbsp;Sorour Alotaibi,&nbsp;Omar Ali,&nbsp;Maria Rashidi,&nbsp;Ahmad Gholami,&nbsp;Qingbin Zheng,&nbsp;Isam Zabalawi,&nbsp;Chun Cheng,&nbsp;Bijan Davvaz","doi":"10.1007/s10462-026-11596-5","DOIUrl":"10.1007/s10462-026-11596-5","url":null,"abstract":"<div><p>Advanced technologies are experiencing significant transformation through AI (Artificial Intelligence), where mathematical foundations serve as the backbone of classical and modern AI models. Despite rapid advancements and widespread adoption, there is a lack of a unified framework for core AI components, comprising building blocks, governing equations, parameters, evaluation criteria, benchmarks, performance metrics, and objective functions across technological domains. In this comprehensive survey, we address this gap in the areas of energy, renewable energy and water, smart buildings and cities, the environment and climate change, hydrogen and hydrogen fuel cells, and cross-sector advanced technologies, including robotics and autonomous systems, computer vision, finance, and industrial manufacturing, with systematic intercomparison and benchmarking. In addition, we conduct a taxonomy-based analysis with a methodological focus on AI models, their underlying parameters and governing equations, as well as their pros, cons, trade-offs, comparative analyses, and directions for future development. This survey consolidates these elements into a structured reference that defines key requirements for AI development in high-tech sectors and provides a forward-looking roadmap to foster innovation beyond current technological infrastructures.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 10","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11596-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148782144","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":"From individual decisions to team emergence: a survey on explainable cooperative multi-agent reinforcement learning","authors":"Lei Sheng,&nbsp;Xiliang Chen,&nbsp;Zhiqiang Pan,&nbsp;Fei Cai,&nbsp;Honghui Chen","doi":"10.1007/s10462-026-11598-3","DOIUrl":"10.1007/s10462-026-11598-3","url":null,"abstract":"<div><p>Multi-Agent Reinforcement Learning (MARL) holds significant promise for cooperative decision-making, yet its reliance on deep neural networks (DNNs) creates “black-box” characteristics that impede trustworthy deployment in high-stakes scenarios. This lack of transparency complicates tracing decision logic and raises concerns about safety and accountability. This survey systematically reviews Explainable MARL (XMARL) for cooperative settings, deconstructing the decision-making chain from individual agent policies to collective team behavior. To address the absence of a unified framework, we introduce a novel multi-level taxonomy encompassing microscopic individual behavior, interaction mechanisms, team strategy emergence, and system-level performance. We organize core explanatory questions and technical approaches within this structure, summarize the principles and limitations of representative methods, and critically discuss key challenges such as evaluation standards, causal reasoning integration, and deployment adaptability. Our goal is to provide both theoretical foundation and technical guidance for building transparent, trustworthy, and verifiably cooperative multi-agent systems (MASs).</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 10","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11598-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148782143","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}