Artificial Intelligence Review最新文献

{"title":"Detecting and mitigating DDoS attacks with AI: a survey","authors":"Alexandru Apostu,&nbsp;Silviu Gheorghe,&nbsp;Andrei Hîji,&nbsp;Nicolae Cleju,&nbsp;Andrei Pătraşcu,&nbsp;Cristian Rusu,&nbsp;Radu Tudor Ionescu,&nbsp;Paul Irofti","doi":"10.1007/s10462-026-11639-x","DOIUrl":"10.1007/s10462-026-11639-x","url":null,"abstract":"<div><p>Distributed Denial of Service attacks represent an active cybersecurity research problem. Recent research shifted from static rule-based defenses towards AI-based detection and mitigation. This comprehensive survey covers several key topics. Preeminently, state-of-the-art AI detection methods are discussed. An in-depth taxonomy based on manual expert hierarchies and an AI-generated dendrogram are provided, thus settling DDoS categorization ambiguities. An important discussion on available datasets follows, covering data format options and their role in training AI detection methods together with adversarial training and examples augmentation. Beyond detection, AI based mitigation techniques are surveyed as well. Finally, multiple open research directions are proposed.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11639-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149189147","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":"A systematic survey of blockchain-enabled artificial intelligence for industrial IoT: recent advances, integration challenges, and future prospects","authors":"Anichur Rahman,&nbsp;Dipanjali Kundu,&nbsp;Tanoy Debnath,&nbsp;Muaz Rahman,&nbsp;Md. Jahidul Islam,&nbsp;Fahim Al Farid,&nbsp;Hezerul Abdul Karim,&nbsp;Abu Saleh Musa Miah","doi":"10.1007/s10462-026-11612-8","DOIUrl":"10.1007/s10462-026-11612-8","url":null,"abstract":"<div><p>In the rapidly evolving landscape of technology, Blockchain (BC), Artificial Intelligence (AI), and Smart Industrial Internet of Things (IIoT) are leading and promising technologies in the world that facilitate the current society to develop the quality of living and make it simpler for users. However, these technologies have been applied in various domains for different purposes. These technologies successfully assist in developing the desired system, such as smart cities, homes, education, and industries. Moreover, these technologies need to address various issues–security, privacy, confidentiality, scalability, and application challenges in diverse fields. Most importantly, while considerable studies have researched the fusion of these technologies, this study stands out considering that it provides a comprehensive survey analysis based on a PRISMA-guided assessment of 295 peer-reviewed studies, focusing on the integration challenges and coordinated benefits of incorporating blockchain and AI into smart IIoT environments. In particular, this survey highlights unexplored research gaps and future directions pertaining to scalability, latency, security, and real-time industrial deployment while offering a structured analytical framework and a methodical organization of Blockchain-enabled AI integration approaches for Industrial IoT. Unlike previous reviews, this survey explores an in-depth analysis of these technologies’ practical applications and security implications, addressing crucial gaps in the existing literature. In this context, with the increasing demand for solutions to these issues, the authors present a detailed survey on the AI approaches with BC in the smart IIoT. Firstly, we focus on state-of-the-art overviews regarding AI, BC, and smart IIoT applications. Then, we identified the benefits of integrating these technologies and discuss the established methods, tools, and strategies efficiently. Most importantly, we highlight the various issues–security, stability, scalability, and confidentiality and guide the way of addressing strategy and methods. Furthermore, the individual and collaborative benefits of applications have been discussed. Lastly, we extensively focused on the open research challenges and potential future guidelines based on BC-based AI approaches in the intelligent IIoT system.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11612-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132585","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":"Balancing fairness and influence spread in social networks: a multi-objective evolutionary approach","authors":"Amjad Osmani,&nbsp;Laleh Tafakori,&nbsp;Mahdi Jalili,&nbsp;Hossein Hassani","doi":"10.1007/s10462-026-11627-1","DOIUrl":"10.1007/s10462-026-11627-1","url":null,"abstract":"<div><p>Influence maximization in social networks has received increasing attention, particularly in applications where fairness among demographic groups is an important concern. However, many existing approaches either overlook group-level disparities or primarily optimize influence spread without explicitly modeling fairness-related trade-offs. In this paper, we propose a group-aware multi-objective evolutionary framework that decomposes seed sets into group-specific sub-solutions. Each demographic group maintains independent subpopulations, which are evolved to improve local influence within the corresponding group. The optimized sub-solutions are then recombined into full solutions, which are globally evaluated using two objectives: total influence spread and inter-group fairness, estimated via pre-generated Reverse Reachable sets. This design reduces the need for repeated diffusion simulations during optimization and can improve computational efficiency. To further enhance diversity and reduce the risk of premature convergence, we introduce a sampling-based strategy that generates new composite solutions by probabilistically selecting best, worst, or random sub-solutions from each group. Multi-objective selection is performed using Pareto dominance, front ranking, an elite archive, and crowding distance. The proposed framework maintains population diversity through subpopulation co-evolution and adaptive sampling. Experimental evaluations conducted on eleven synthetic networks and five real-world datasets show that the proposed method achieves competitive or improved performance relative to strong baselines, considering both fairness of coverage across groups and overall influence propagation.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11627-1.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132726","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-governed hospitals-of-the-future under industry 5.0: intelligent personalisation, cloud-integrated AI, and human-centred governance","authors":"Amr Adel,&nbsp;Mohammad Al-Rawi,&nbsp;Anna Shillabeer,&nbsp;Patrick Shearman,&nbsp;Rouwa Yalda,&nbsp;Tony Jan,&nbsp;Asmaa Soliman Al-Moisher,&nbsp;Mohammad Ali Moni","doi":"10.1007/s10462-026-11646-y","DOIUrl":"10.1007/s10462-026-11646-y","url":null,"abstract":"<div><p>Artificial intelligence is reshaping hospital care delivery through federated learning pipelines, edge-cloud inference architectures, and AI-driven clinical decision support. Yet the translation of these AI capabilities into patient-centred, institutionally governable, and humanised hospital systems remains fragmented across the literature. This paper addresses that gap through a PRISMA-compliant systematic evidence synthesis of 116 included studies and reports (inter-rater reliability <span>(kappa = 0.884)</span>, 95 % confidence interval (CI): 0.835−0.933), and is careful to separate its empirical findings from its conceptual contribution. Empirically, we synthesise evidence at the level of three individual AI-enabled pillars: personalisation (AI-driven genomic pipelines, AI-assisted diagnostics using convolutional neural network (CNN) and transformer architectures, continuous physiological monitoring, and longitudinal predictive models); cloudisation (multi-platform cloud architectures evaluated via a weighted multi-criteria decision analysis (MCDA), Fast Healthcare Interoperability Resources (FHIR)-native data lakes, federated AI health monitoring, and data lineage, provenance auditing, and role-based access control); and humanisation (empathy-driven interfaces, co-design evaluation frameworks, digital therapeutics regulation, and operational ethical AI governance with enforceable accountability). Conceptually, we propose that these pillars are co-constitutive and, under an Industry 5.0 governance frame, jointly define the Hospital-of-the-Future; we advance this integrated tri-pillar framework as an analytical proposition rather than an empirically validated system. The pillar-level evidence is substantial: meta-analytic standardised mean differences (SMDs) of 0.48−0.64 for personalisation outcomes, 82.4 % sensitivity and 91.1 % specificity for AI-integrated sepsis screening (<span>(n = 5{,}765)</span>), and a 0.7-day length-of-stay reduction (<span>(p = 0.031)</span>) for humanised ward design. The evidence base is, however, fragmented across the pillars: no included study evaluated the three pillars jointly, and although individual technology domains span Technology Readiness Levels (TRL) 3–8, the integrated framework itself remains at TRL 3–4. A cluster randomised trial protocol is specified to generate the missing integration evidence and to advance the framework toward TRL 6–7. The review therefore offers a promising, evidence-informed conceptual architecture and a testable research agenda for AI-enabled hospitals that are scalable, governable, and human-centred, while making explicit the gap between the currently fragmented evidence and the integrated vision it proposes.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11646-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132725","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":"The interpretability paradox in cancer imaging and risk prediction: a critical narrative review of explainable AI, failure modes, and design alternatives","authors":"Omer S. Alkhnbashi,&nbsp;Rasheed Mohammad,&nbsp;Costerwell Khyriem,&nbsp;Sven Hauns,&nbsp;Ankita Singh,&nbsp;Imran Khan,&nbsp;Roshna Lawrence Gomez,&nbsp;Revathy Ramachandran,&nbsp;Moaza Hashim AlBedwawi,&nbsp;Manal Mohammed AbdulRahim,&nbsp;Zaid AbdelAziz,&nbsp;Ali. M. Batarfi,&nbsp;Ahmad Abou Tayoun,&nbsp;Nelson C. Soares,&nbsp;Mohammed Uddin,&nbsp;Babacar Cisse,&nbsp;Fahad Ali,&nbsp;Rolf Backofen","doi":"10.1007/s10462-026-11634-2","DOIUrl":"10.1007/s10462-026-11634-2","url":null,"abstract":"<div><p>Deep learning has advanced cancer imaging and cancer-related risk prediction, but many high-performing models remain difficult to interrogate in clinically meaningful terms. This creates an interpretability paradox: gains in predictive performance often coincide with reduced transparency, while widely used post-hoc explanations can be persuasive without providing reliable evidence of model reasoning. Here, we present a critical narrative review and position argument, supported by a semi-systematic mapping of the 2017–2025 literature across major biomedical and technical databases, to reassess explainability practices in oncology and identify design alternatives that are better aligned with clinical oversight. Across cancer applications, explainability is most operationalised through post-hoc attribution methods, including saliency-style heatmaps for imaging and local feature attribution for structured risk models, with evaluation frequently limited to plausibility narratives or internal technical proxies. We argue that this practice leaves an evaluation gap: limited evidence on explanation fidelity, stability, calibration-aware interpretation of risk outputs, and impact on clinician decision-making under realistic workflow constraints and distribution shift. To address this gap, we delineate three underused but clinically aligned design families: concept-based modelling, counterfactual and contrastive reasoning, and prototype- or case-based approaches, alongside emerging hybrid and interpretable-by-design architectures. We synthesise these insights into a pragmatic lifecycle framework in which post-hoc Explainable AI (XAI) is treated primarily as a diagnostic tool for critique, auditing, and monitoring, while transparency-by-design is evaluated as a primary modelling target for high-stakes use. We conclude with a research and governance agenda centred on domain-grounded concept resources, clinically feasible counterfactuals with validity checks, curated exemplar libraries, and prospective human-in-the-loop evaluation of transparency as an intervention.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11634-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132241","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":"Social cognitive architecture for NPC groups: integration of transformer theory of mind and hierarchical reinforcement learning","authors":"Zhenyu Qiu,&nbsp;Xiaojun Tang","doi":"10.1007/s10462-026-11630-6","DOIUrl":"10.1007/s10462-026-11630-6","url":null,"abstract":"<div><p>Non-Player Characters (NPCs) require social cognition to enable intelligent and interactive behaviours within virtual environments. In gaming and other multi-agent systems, current NPC models often fall short in social intelligence and coordination because they cannot infer or anticipate the mental states of other agents. To address this gap, this paper introduces a novel social cognitive architecture that integrates Hierarchical Reinforcement Learning (HRL) with a Transformer-based Theory of Mind (ToM). The framework enables NPCs to make goal-oriented decisions while simultaneously inferring and adapting to the beliefs, goals, and intentions of others. By combining the social awareness capabilities of the ToM module with the strategic decision-making strength of HRL, NPCs are able to coordinate actions more effectively and respond dynamically to evolving scenarios. On a held-out test split of the MCPDial dialogue dataset, the proposed architecture achieves 96.33% accuracy in predicting NPCs’ next action and inferred mental state from dialogue context, a task-specific measure of next-turn prediction performance rather than a general-purpose Theory-of-Mind benchmark, representing a 5.35% improvement over baseline HRL models without ToM (91.96%) on the same task Further enhancements are observed in agent coordination and dialogue coherence. Overall, this approach advances both task-driven reasoning and social interaction in complex multi-agent environments, marking a significant step toward more intelligent and socially aware NPCs.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11630-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132240","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":"Bridging Modalities: A Survey and Taxonomy of Multimodal AutoML (MMAutoML)","authors":"Blaž Škrlj,&nbsp;Aljaž Osojnik","doi":"10.1007/s10462-026-11624-4","DOIUrl":"10.1007/s10462-026-11624-4","url":null,"abstract":"<div><p>Automated machine learning (AutoML) reduces the cost of developing high-performing models by automating pipeline design, model selection, and hyperparameter optimization. As data sets increasingly combine heterogeneous sources (e.g., text, images, audio, and structured/tabular records), AutoML is extending to multimodal settings, where performance depends on representation learning and cross-modal fusion as well as model search. This survey reviews automated multimodal machine learning (MMAutoML), distinguishing core systems that automate both representation and fusion decisions from near-core systems, partial multimodal AutoML tools, and adjacent multimodal ML infrastructure. We organize multimodal learning around early, late, and hybrid fusion paradigms and discuss trade-offs in robustness, interpretability, and deployment. We then curate and compare representative open-source and commercial MMAutoML frameworks, summarize supported modalities, and characterize the ecosystem via time-evolution and similarity-based groupings. Finally, we overview applications in healthcare, autonomous systems, finance, and e-commerce, and highlight open challenges in modality alignment, missing or degraded inputs, efficiency and scalability, reproducibility, and governance (privacy, bias, and monitoring). We conclude with practical guidance and research directions toward reliable, end-to-end MMAutoML.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11624-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149189283","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":"Symmetry-aware learning in machine intelligence: architectural principles and deployment trade-offs","authors":"Maria Trigka,&nbsp;Elias Dritsas","doi":"10.1007/s10462-026-11631-5","DOIUrl":"10.1007/s10462-026-11631-5","url":null,"abstract":"<div><p>Symmetry-aware machine learning (ML) embeds invariance and equivariance constraints directly into model architectures, providing principled inductive biases that can improve generalization, sample efficiency, and alignment with data-generating processes. This survey presents a principled synthesis of symmetry-aware architectures, treating symmetry as the core architectural design axis, rather than as an auxiliary modeling property. We introduce a structured taxonomy of symmetry types by mapping discrete, continuous, and statistical group actions to their architectural realizations, including group-equivariant convolutions, equivariant graph networks and attention mechanisms, neural operators, and symmetry-constrained generative models. Beyond architectural categorization, the survey examines how symmetry constraints shape the representation capacity, training stability, robustness, and computational behavior. Representative deployment scenarios are analyzed across molecular modeling, scientific ML, three-dimensional (3D) vision, and relational reasoning, with explicit attention to symmetry-aligned benchmarks, evaluation protocols, performance regimes, inference costs, hardware compatibility, and interactions with low-precision quantization. In addition to conceptual synthesis, the survey consolidates representative empirical indicators reported in the literature, including benchmark outcomes and computational complexity characteristics across major symmetry-aware-architecture families. Across multiple domains, reported studies frequently indicate improved sample efficiency, robustness under symmetry-preserving transformations, and reduced prediction error compared with unconstrained models when symmetry assumptions align with the underlying data-generating processes. In contrast to prior studies that emphasize abstract formalism or isolated domains, this work integrates theoretical foundations, architectural mechanisms, empirical evidence, and system-level considerations into a unified analytical framework, highlighting emerging directions in scalable symmetry-aware learning and hardware-aware co-design.</p></div>","PeriodicalId":8449,"journal":{"name":"Artificial Intelligence Review","volume":"59 11","pages":""},"PeriodicalIF":18.8,"publicationDate":"2026-07-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10462-026-11631-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132034","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}