Cancer is a major and growing global health challenge, with almost 20 million new cases and over 10 million deaths annually. Projections suggest that both incidence and mortality will rise by 2050, driven by global aging, demographic shifts, and changes in risk factors. Against this backdrop, this paper studies the increasing complexity of oncology therapeutic decision-making as multimodal clinical and molecular datasets expand. To address these challenges, a Multimodal Cross-Attention Transformer framework is proposed that integrates structured electronic health records and unstructured narrative reports. Specifically, it uses specialized encoders for tabular and textual data and fuses them via a cross-attention mechanism. This system aims to improve treatment recommendations and maintain clinician interpretability by leveraging explainable artificial intelligence (XAI), addressing current constraints in oncology decision support systems. Additionally, it preserves modality-specific information, captures complex relationships among data types, and integrates the model's reasoning with clinical concepts. Ultimately, the system aims to improve personalized treatment selection and support healthcare workflows, particularly in real-world hospital settings with heterogeneous, incomplete data.

This work proposes a novel neuro-symbolic multi-agent coordination framework that integrates Deep Reinforce-ment Learning (DRL) with a Market-Driven Hierarchical Neuro-Fuzzy Politree architecture. The proposed model, termed DRL-MD-HNFP, addresses key limitations of traditional tabular reinforcement learning in scalable multi-agent environments, including limited eneralization, high-dimensional state-action spaces, and coordination inefficiencies. The rchitecture preserves the interpretability and modularity of hierarchical fuzzy role allocation while eplacing SARSA-based value estimation with deep neural approximators, enabling robust policy earning in complex and partially observable domains. A decentralized auction-based coordination mechanism dynamically allocates tasks according to adaptive cost functions, promoting efficient resource distribution and role specialization without centralized control. The model is validated in benchmark pursuit game scenarios under multiple grid sizes and initializa-tion configurations. Experimental results demonstrate substantial performance gains, achieving a 45.8% reduction in average capture steps on a 9×9 grid and a 52.7% reduction on a 99×99 grid when compared to classical MA-RL-HNFP baselines. These improvements confirm faster convergence, enhanced scalability, and superior coordination efficiency. These findings establish the DRL-MD-HNFP as a scalable and interpretable neuro-symbolic framework for next-generation multi-agent systems operating in dynamic and resource-constrained environments.

Projects developing advanced digital solutions for industry are characterized by high technical complexity, multiple stakeholders, and ever-changing expectations. In such contexts, project success depends not only on technical performance but also on the partners' ability to remain aligned throughout the development and implementation phases. This article analyzes the DiMAT Horizon Europe project, which involved 18 organizations, 9 digital toolkits, and 4 industrial pilots, with the aim of identifying the main critical issues that emerged during implementation and translating them into methodological insights for future collaborative projects. The study is based on a lessons learned analysis conducted at the end of the project through qualitative and quantitative questionnaires addressed to pilot projects and toolkit developers. The results show that the most significant challenge concerned initial alignment, particularly regarding the initial clarity of the toolkits' functionalities, requirements, and expected value. Data availability and quality also emerged as a significant challenge, often amplifying misalignments and requiring adjustments in toolkit implementation. Feedback from external early adopters further confirms that these issues reflect broader challenges of adoption and contextual alignment. Based on these findings, the article proposes a trigger-based methodology for continuous alignment, designed to support the early identification of critical issues and the definition of corrective actions in future collaborative innovation projects.

High-stakes banking journeys hinge on brief “critical moments” where customers face heightened vulnerability, time pressure, and limited control. In regulated contexts, trust can diverge between what actors experience in the interaction (perceived trust, PT) and what is justified given the robustness, governance, and accountability of service delivery (warranted trust, WT). This paper reports exploratory, theory-building qualitative research based on semi-structured interviews and Critical Incident Technique (CIT) [2], [3] style incident distillation across four banking journeys in a large Nordic bank: private banking/family office services, retail mortgage credit, corporate lending/financing, and corporate onboarding/KYC. We map ten representative incidents into a 2×2 PT×WT space grounded in Mayer–Davis–Schoorman’s trust model and ABI dimensions (ability, benevolence, integrity) [1]. Findings reveal two dominant misalignment trajectories: (1) communication breakdowns that erode PT under comparatively stable WT (undertrust), and (2) execution and handoff breakdowns that degrade both PT and WT (justified distrust). We contribute a practical diagnostic—a PT×WT hero map supported by an audit-trail table and an incident-card template—that helps managers locate trust breakdowns in distributed service chains and select repair levers. We discuss how the framework supports responsible LLM adoption: LLMs can improve PT through fluent explanations and responsiveness, but only governance-by-design and operational resilience can sustain WT and prevent trust miscalibration.

With the rapid advancement of technology, the smartphone market has gradually matured, prompting major technology companies to embark on a new round of competition. Among these emerging fields, wearable technology has become a focal point for the latest generation. Smartwatches, as the most representative category of wearable technologies, are characterized by their rapid evolution, multifaceted technical architecture, and increasing market penetration. Given the intensifying global competition and the accelerating pace of technological development, it is essential to analyze the current technological landscape and to forecast emerging trends. This study therefore seeks to analyze the developmental trajectory and technological focus areas of smartwatch innovations and to predict future technological trends in the smartwatch industry through patent-based trend analysis, thereby providing empirical evidence to guide industry stakeholders and future researchers.