ABSTRACT Cross‐modal information systems have emerged as an important area of research that enables information retrieval in different modalities such as text, image and audio. Existing Cross‐modal retrieval methods focus on semantic consistency between modalities but rarely consider retrieval effectiveness in generative models. This study investigates the effectiveness of the Multimodal Generative Advertisement Network (MMGAN) in enhancing Cross‐modal Information Retrieval (CMIR) performance. Using benchmark datasets that include various modalities, it conducts a comprehensive analysis of MMGAN against several state‐of‐the‐art models, including traditional Generative Adversarial Networks (GAN) and Convolutional Neural Networks (CNN). The experimental evaluation considers both representation quality and cross‐modal retrieval effectiveness. Controlled unimodal benchmarks are used to examine latent representation quality, whereas MS‐COCO, containing paired image–text samples, serves as the primary benchmark for genuine cross‐modal retrieval. Retrieval performance is assessed using Mean Average Precision (mAP), Precision@K, Recall@K, and NDCG@K for image‐to‐text and text‐to‐image retrieval, while NMI, ARI, and clustering accuracy are employed as supplementary measures of representation and semantic grouping quality. The results demonstrate the effectiveness of the proposed MMGAN‐based framework in learning semantically aligned multimodal representations and improving cross‐modal retrieval performance. Furthermore, it analyses the strengths and limitations of MMGAN, informing readers of their potential applications and exploring future research directions in CMIR. It contributes to the advancement of generative models to improve Cross‐modal retrieval systems, emphasizing more efficient and effective information retrieval.
ABSTRACT The rapid proliferation of latency‐sensitive and data‐intensive applications has accelerated the adoption of cloud–fog computing architectures to support real‐time processing and distributed resource management. However, efficient task scheduling in cloud–fog environments remains a challenging problem because of resource heterogeneity, dynamic workloads, network variability, and stringent quality of service (QoS) requirements. Artificial intelligence (AI)‐based techniques, including machine learning (ML), deep learning (DL), deep reinforcement learning (DRL), and hybrid AI approaches, have emerged as promising solutions for addressing these challenges. This paper presents a systematic review of AI‐based task scheduling techniques in cloud–fog computing environments published between 2021 and 2025. Following a PRISMA‐inspired methodology, 74 high‐quality studies were selected and analyzed. The reviewed works are classified into ML, DL, DRL, and hybrid AI‐based scheduling categories and evaluated with respect to scheduling objectives, QoS parameters, datasets, evaluation tools, and implementation environments. The analysis reveals that AI‐driven scheduling approaches significantly improve key performance metrics, including task scheduling performance (14%–90%), latency reduction (15%–65%), energy optimization (15%–41%), load balancing (20%–45%), execution time reduction (15%–45%), and cost optimization (10%–32%). This review further highlights the increasing adoption of DL, DRL, and hybrid AI techniques for managing complex and dynamic cloud–fog infrastructures. Additionally, this study identifies critical research challenges, including scalability, resource heterogeneity, lack of standardized benchmark datasets, limited real‐world deployments, and insufficient consideration of network‐layer dynamics. Finally, emerging research directions and future opportunities for developing scalable, adaptive, and intelligent scheduling frameworks in cloud–fog computing environments are discussed. This article is categorized under: Technologies > Cloud Computing Technologies > Internet of Things Technologies > Artificial Intelligence
ABSTRACT Artificial intelligence (AI) has rapidly transformed smart surveillance systems, significantly leveraging technologies such as deep learning, computer vision, and adaptive monitoring. Driven by the need for real‐time object detection, behavioral analysis, and automated alerting, AI‐powered surveillance is increasingly vital for enhancing public safety, operational efficiency, and decision‐making in fields such as security, health‐care, transportation, and environmental monitoring. The growing societal impact and complexity of surveillance necessitate a comprehensive examination of foundational methodologies, key application domains, and the technical, ethical, and operational challenges associated with these systems. Analysis reveals that although AI‐driven surveillance solutions offer improved accuracy, scalability, and responsiveness, they continue to face notable barriers including high computational demands, privacy and data quality concerns, interpretability limitations, and inadequate regulatory frameworks. The review further identifies critical emerging trends, such as sustainable AI practices, human‐AI collaborative models, and the integration of AI with the Internet of Things (IoT) and blockchain technologies, as significant influences shaping the future trajectory of the field. Addressing these challenges through ongoing research, innovation, and responsible deployment is essential to fully unlock the societal and industrial potential of AI‐enabled surveillance solutions.
ABSTRACT Multimodal artificial intelligence (AI) increasingly integrates imaging, pathology, omics, electronic health records, clinical text, and physiological signals. However, multimodality is often defined by the number of input types, leaving unclear when additional data improve a clinical decision, duplicate available evidence, or introduce informative disagreement. This concept‐driven critical perspective proposes a relationship‐centered framework for clinically meaningful multimodality. Drawing on a focused synthesis of recent PubMed literature and foundational studies, we define complementarity, redundancy, and conflict as context‐dependent relationships shaped by the clinical question, decision point, timing, and available evidence. We examine how attention mechanisms, gating mechanisms, graph structures, mixture‐of‐experts models, and modular encoders can represent or regulate these relationships. We then organize the framework across five dimensions linking modality relationships to clinical meaning, representative contexts, modeling strategies, and evaluation approaches. Relationship‐centered design principles are further connected to a staged evaluation pathway spanning representation‐level validation, relationship‐aware technical validation, clinical‐utility assessment, prospective implementation, and monitoring. Pulmonary nodule assessment, reperfusion decision‐making in acute ischemic stroke, and immunotherapy response assessment illustrate how the same modality may shift between complementary, redundant, and conflicting roles across clinical workflows. We argue that multimodal systems should be evaluated not only by discrimination, but also by marginal modality value, calibration, reduced‐input robustness, behavior under missing, asynchronous, or discordant evidence, and effects on clinical decisions. Future research should prioritize relationship‐preserving representations, decision‐aware modality acquisition, relationship‐specific benchmarks, and human‐centered evaluation. Clinically meaningful multimodality depends not on how many data streams a model can fuse, but on whether it uses each stream for a defined clinical purpose while preserving evidence provenance, timing, and uncertainty.
ABSTRACT Among the main objectives of recommender systems, the minimization of the problem known as information overload stands out. The personalization of recommendations, which emerge as users interact across various digital platforms, aims to mitigate this recurring issue in the modern world. Over the years, different recommendation paradigms have been explored, revealing both the potentialities and limitations of each technique and their respective proposed models. As a result, it has been shown that combining two recommendation paradigms, that is, the hybridization of methodologies, has the potential to be a more efficient alternative to meet the expectations and needs of these systems. Accordingly, this study presents a systematic literature review on recommender systems, focusing on two paradigms: context‐aware recommendation and session‐aware recommendation (CASARS). Our research seeks to elucidate which studies in the literature address recommendation strategies that leverage users' historical information to contribute to item recommendations throughout their ongoing sessions, as well as mechanisms for incorporating situational contextual information, that is, the user's context expressed strictly through interactions among items within the current session. Our goal is to outline the trajectory of the proposed models, identify the predominant application domains, critically analyze the evaluation methodologies, and, based on that, highlight the fundamental challenges and future research directions. In this study, we adopted a rigorous systematic review protocol, through which 25 articles were selected and thoroughly analyzed. The analysis reveals a clear chronological and conceptual progression of architectures, evolving from sequential models based on Recurrent Neural Networks to more complex structural representations with Graph Neural Networks, culminating in current approaches employing Transformers and attention mechanisms. We identified the main application focuses in domains such as e‐commerce, which accounts for 44% of the studies, and highlighted a critical gap in evaluation practices, characterized by the preference for accuracy metrics, such as Recall and NDCG, over dimensions essential to user experience, such as diversity, novelty, and serendipity. Although the field demonstrates rapid evolution and technical sophistication, future research directions were identified and thoroughly discussed.
