Art Institute of Portland

This paper presents a case study of an artificial intelligence (AI)-supported media monitoring system implemented in a European commercial banking group to enhance early detection of emerging risks. The pipeline processes millions of media articles daily using a scalable Databricks-based architecture and large language models (LLMs) for relevance scoring, novelty detection and summarisation. It demonstrates how AI-based text analysis supports credit, liquidity and geopolitical risk monitoring by transforming unstructured news into risk-relevant intelligence delivered through automated e-mail briefings, dashboards and early warning system (EWS) integration. The case study further illustrates how such AI-supported media monitoring can be governed and deployed within the regulatory and supervisory framework of the European banking sector while strengthening risk awareness and decision making and offering additional value for compliance, operational risk and communication functions. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.

Evidence-based practice is a critical skill for nurses and health care professionals. Because interprofessional teams are based in a culture of clinical inquiry, they must ask important practice questions, search for and appraise relevant evidence, and, based on the best available evidence, design single, bundle, or blended interventions to improve targeted outcomes. The Consolidated Framework for Implementation Research is an implementation science model that teams can use to not only plan these interventions but also implement and evaluate selected interventions in the local practice setting. This article, as part 1 of a series, will review considerations for designing interventions using a collection of Critical Care Nurse articles on hospital-acquired pressure injury prevention initiatives as exemplars. In part 2, I will explore the process of planning and executing the most effective strategies to embed the evidence-based intervention into practice in a sustainable way.

Sporadic-E (Es) layers occur irregularly and show strong seasonal concentration. At mid-latitudes, occurrence rates are high in summer and low in the rest of the year. Such seasonal bias can cause machine-learning predictors to overpredict in summer (May–Aug) and underpredict in non-summer (Sep–Apr). Using Jeju ionosonde data for 2009–2024, we compare three strategies to mitigate this bias: (1) monthly oversampling, (2) threshold adjustment, and (3) probability calibration. Performance is evaluated with the true skill statistic (TSS), geometric mean (G-mean), and count-based symmetry index (C-SI). Experiments show that combining monthly oversampling with threshold adjustment yields the most stable and consistent improvements, significantly reducing summer overprediction and non-summer underprediction. Shapley additive explanations (SHAP) analysis indicates that foEs (the Es critical frequency), day of year (DOY), and vertical ion convergence (VIC) are the key predictors in the model, consistent with known Es formation mechanisms. This study underscores the importance of training strategies that account for seasonal data bias in mid-latitude Es prediction and points to future multi-station extension and model generalization.

The autonomous business decision systems are now the centre of organisational competitiveness as organisations continue to work in volatile and increasingly data-intensive and interdependent environments. The use of multi-agent systems (MAS) can provide an effective paradigm, which facilitates decentralised, adaptive and collaborative intelligence with distributed agents perceiving environments and negotiating actions and autonomous optimisation. This paper is a critical assessment of how multi-agent AI models, such as logic agents, multi-agent reinforcement learning (MARL), agents based on large language models, agent development systems, simulation systems, and self-evolving agents' systems, have contributed to autonomous business decision systems. Systematic literature reviews, simulations, enterprise case studies and foundation-model based research and decision intelligence (DI) frameworks evidence shows how multi agent strategies provides increased robustness, scalability and responsiveness to the complex organisational domains. Simulation-based research using MAS is seen to be able to model emerging behaviours, test organisational conditions and uncertainty within energy system, construction, logistics and financial fields. The further reinforcement learning and cooperative decision modelling enhance MAS autonomy as it provides the ability to optimise the learning of dynamic environments. Similar results are presented by the research of Automated Machine Learning (AutoML) that demonstrates that automated model development helps lower technical barriers and speeds up its deployment and enables adaptive decision pipelines. Decision Intelligence literature displays how organisations can restructure decision-making with the help of AI to construct data pipelines, and provide human-computer interaction. Although issues remain, such as explainability, cost of coordination and data governance, there is strong evidence that MAS constitute a backbone architecture to next generation autonomous decision ecosystems according to all sources. The given synthesis illustrates how multi-agent AI may be used as a strategic foundation of predictive, operational and strategic automation of decisions in contemporary businesses.

Ammonia (NH3) is considered a key fuel for future carbon-neutral energy systems. Understanding its combustion characteristics under intense turbulence with high Karlovitz (Ka) numbers is crucial for industrial applications. In this study, a three-dimensional quasi-direct numerical simulation (quasi-DNS) is performed on an experimental premixed ammonia jet flame at a Ka number of 1643. The simulation is conducted using the open-source computational platform DeepFlame, employing a minimum grid resolution of 56 μm and second-order spatial and temporal discretization schemes. The chemical system is modeled using a semi-detailed mechanism comprising 26 species and 100 elementary reactions. The simulation results show good agreement with laser-based diagnostic measurements. The general flame structure obtained from the simulation is consistent with fundamental combustion theory. In particular, turbulent eddies are observed to penetrate and broaden the reaction zone, with this effect intensifying downstream. Based on these results, turbulence–chemistry interactions and NO formation characteristics are systematically analyzed. For turbulence–chemistry interactions, the flame surface density transport theory is employed to clarify the roles of tangential strain rate, curvature stretch, and convection in governing the flame surface density distribution. Regarding NO formation, NO is primarily generated through the HNO and thermal pathways and consumed via the N2O and NHi pathways, leading to a net positive production on the product side of the flame. Furthermore, the influence of progress variable definitions is demonstrated in both turbulence–chemistry interactions and NO formation. It is found that the H2O-based progress variable captures more comprehensive information about the spatial and temporal distribution of chemical processes. These findings provide valuable insights into ammonia combustion under high-Ka conditions and contribute to the development of high-fidelity simulation databases essential for future modeling efforts.

Human motion video generation has garnered significant research interest due to its broad applications, enabling innovations such as photorealistic singing heads or dynamic avatars that seamlessly dance to music. However, existing surveys in this field focus on individual methods, lacking a comprehensive overview of the entire generative process. This paper addresses this gap by providing an in-depth survey of human motion video generation, encompassing over ten sub-tasks, and detailing the five key phases of the generation process: input, motion planning, motion video generation, refinement, and output. Notably, this is the first survey that discusses the potential of large language models in enhancing human motion video generation. Our survey reviews the latest developments and technological trends in human motion video generation across three primary modalities: vision, text, and audio. By covering over two hundred papers, we offer a thorough overview of the field and highlight milestone works that have driven significant technological breakthroughs. Our goal for this survey is to unveil the prospects of human motion video generation and serve as a valuable resource for advancing the comprehensive applications of digital humans.

In recent years, the rapid advancements in computer science have spurred the development of various cutting-edge intelligent algorithms. Among these, the transformer, which is built upon a multi-head attention mechanism, is one of the most prominent AI models. The advent of such algorithms has significantly advanced retrosynthesis prediction, though challenges remain in chemical interpretability and real-world deployment. Unlike traditional models, AI-based retrosynthesis prediction systems can automatically extract chemical knowledge from vast datasets to forecast retrosynthesis pathways. This review provides a comprehensive overview of modern intelligent algorithms applied to retrosynthesis prediction, with a particular focus on artificial intelligence techniques. We begin by discussing key deep learning models, then explore available chemical reaction datasets and molecular representations. The discussion extends to the latest state-of-the art in AI-assisted retrosynthesis models, including template-based, template-free, and semi-template-based approaches. Finally, we compare these models across various classifications, highlighting several challenges and limitations of current methods, and suggesting promising directions for future research.

Large Language Models (LLMs) have shown remarkable capabilities as AI agents. However, existing methods for enhancing LLM-agent abilities often lack a focus on data quality, leading to inefficiencies and suboptimal results in both fine-tuning and prompt engineering. To address this issue, we introduce EDGE, a novel approach for identifying informative samples without needing golden answers. We propose the Guideline Effectiveness (GE) metric, which selects challenging samples by measuring the impact of human-provided guidelines in multi-turn interaction tasks. A low GE score indicates that the human expertise required for a sample is missing from the guideline, making the sample more informative. By selecting samples with low GE scores, we can improve the efficiency and outcomes of both prompt engineering and fine-tuning processes for LLMs. Extensive experiments validate the performance of our method. Our method achieves competitive results on the HotpotQA and WebShop and datasets, requiring 75% and 50% less data, respectively, while outperforming existing methods. We also provide a fresh perspective on the data quality of LLM-agent fine-tuning.

The EL-MIATTs (Evaluation and Learning with Multiple Inaccurate True Targets) framework is a novel scientific paradigm, developed based on the core principles of LAF (Logical Assessment Formula) and UTTL (Undefinable True Target Learning), aimed at tackling situations where acquiring the true target for a machine learning task is difficult, expensive, or fundamentally impossible. In this paper, bridging theory and practice, we present a case study applying the EL-MIATTs framework to the bicycle lane segmentation task in street images—a task chosen due to its inherent ambiguity and challenges in consistently defining and delineating bicycle lanes in complex urban scenes. We demonstrate that by adhering to the core principles of LAF and UTTL, an appropriate implementation of the EL-MIATTs framework can generate an optimized predictive model that uncovers the underlying true target for a given input street image, achieved through evaluation and learning that leverage the distributional representation of diverse properties encoded within the MIATTs corresponding to the true target underlying that input. Our results highlight the framework’s potential to effectively handle scenarios where obtaining accurate true targets is challenging, offering a scientific and adaptable approach for real-world tasks. This case study thus validates the feasibility of translating the EL-MIATTs theoretical paradigm into practical applications, paving the way for further exploration and broader adoption in related domains.