An intelligent and agile traffic signal system has emerged as a vital sustainable component of urban mobility. The centralised traffic control systems currently in use are not capable of providing the required responsiveness or scalability to facilitate real-time traffic management. In this article, we propose a lightweight hybrid system that integrates a Gated Recurrent Unit (GRU) based predictive model and Digital Twin (DT) technology to provide decentralised, real-time traffic signal optimisation. The GRU model forecasts future localised congestion events from vehicle based Internet of Things (IoT) sensors, while the DT model ensures adequate performance by validating and adjusting control actions based on live information concerning roadway condition changes. We present results that demonstrate improvement of predictive accuracy by 33% (mean absolute error = 4.5), control latency of 78 ms, and a 15% decrease in CO 2 emissions, along with substantial decreases in both congestion and travel time. It was found that the proposed model consistently outperformed the state of the art solutions with improved prediction, latency, and environmental efficiency. The proposed architecture provides superior real-time traffic management within smart city environment.
Large language models (LLMs) and search engines have the potential to perpetuate biases and stereotypes by amplifying existing prejudices in their training data and algorithmic processes, thereby influencing public perception and decision-making. While most work has focused on Western-centric AI technologies, we examine social biases embedded in prominent Chinese-based commercial tools, the main search engine Baidu and two leading LLMs, Ernie and Qwen. Leveraging a dataset of 240 social groups across 13 categories describing Chinese society, we collect over 30 k views encoded in the aforementioned tools by prompting them to generate candidate words describing these groups. We find that language models exhibit a broader range of embedded views compared to the search engine, although Baidu and Qwen generate negative content more often than Ernie. We also observe a moderate prevalence of stereotypes embedded in the language models, many of which potentially promote offensive or derogatory views. Our work highlights the importance of prioritizing fairness and inclusivity in AI technologies from a global perspective.
Banana leaf diseases pose a significant global threat to agricultural productivity and economic stability, substantially reducing the quality and quantity of yield. Given the critical role of banana leaves in the overall growth and development of banana plants, their susceptibility to a wide range of diseases represents a pressing concern. This review systematically explores recent advancements in diagnosing and classifying banana leaf diseases through Artificial Intelligence (AI)-based techniques. Key methodologies reviewed include image preprocessing, machine learning, deep learning, and transfer learning. Particular emphasis is placed on lightweight deep learning architectures, which offer the advantages of high diagnostic accuracy, rapid processing, and minimal computational requirements, making them suitable for deployment in resource-constrained environments. The presence of numerous banana cultivars, each exhibiting subtle variations in leaf morphology and pigmentation, further complicates the detection process, underscoring the need for adaptable and robust AI models. The review also highlights data acquisition, preprocessing strategies, and dataset weaknesses, along with evaluation metrics used to assess model performance. Finally, it identifies existing challenges and research gaps in current approaches with the brief case study by synthesizing these insights. The review provides a comprehensive understanding of AI-powered solutions for the effective detection and classification of banana leaf diseases and their potential practical applications in precision agriculture.
Artificial Intelligence (AI) has transformed the development of game agents, providing new levels of interactivity and engagement. Real-time decision-making, as in fighting games, abets the need for adaptive and human-like behaviour for agents, making competition difficult. In the classics of fighting games, traditional AI is based on pre-programmed scripts, rules-based systems, or approaches that are easily predictable and provide less engaging gameplay. This paper presents an Adaptive AI based on Particle Swarm Optimization (PSO) to adapt its strategies dynamically based on the opponent’s behaviour. The proposed approach enables constant real-time updates to neural network weights, thus making continuous learning, strategic adaptation, and variation in gameplay. The proposed AI is evaluated against multiple state-of-the-art AI models and human players with several performance metrics like Élő Rating, Glicko-2, opponent adaptation score, engagement score, and win consistency score. Experimental results show that the proposed Adaptive AI performs better than other AI in terms of its adaptability, strategic diversity, engagement, and the level of competitiveness it provides against human opponents, which is fair and challenging at the same time. From the findings, it is concluded that real-time optimization can be achieved by integrating PSO with neural networks, which helps improve capabilities in fighting games. The research brings value to the field by creating an adaptable AI agent that enhances user gameplay.
This article presents a novel decision-support framework, Hybrid AI-Augmented Decision Optimization (HAI-HDM), designed to accelerate and improve technology adoption in small and medium enterprises (SMEs). HAI-HDM bridges artificial intelligence and human expertise to deliver context-aware, data-driven technology recommendations custom-made to the unique challenges of SMEs. The framework has five core components: data acquisition and preprocessing, artificial intelligence (AI)-powered technology ranking, human-AI decision integration, explainable recommendation generation, and adaptive learning. To support analytical insights, HAI-HDM utilizes machine learning algorithms. For transparency and confidence, it integrates explainable AI (XAI) tools SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), making the foundation behind each recommendation interpretable. A key feature is its dynamic weighting mechanism, which adjusts the stimulus of human judgment and AI confidence based on the degree of vagueness and corporate context. The framework also employs reinforcement learning, leveraging feedback from real-world adoption situations to continuously improve its recommendations. A case study focused on cloud technology adoption establishes the practical effectiveness of HAI-HDM, showing how it aligns technological choices with professional constraints and strategic goals. By combining analytical power with human insight, the framework not only supports informed decision-making but also promotes greater trust and accountability in AI-driven strategic planning.
Sanjana Singamsetty, Swetha Ghanta, Sujit Biswas
et al.
Chronic renal disease (CRD) is a significant concern in the field of healthcare, highlighting the crucial need of early and accurate prediction in order to provide prompt treatments and enhance patient outcomes. This article presents an end-to-end predictive model for the binary classification of CRD in healthcare, addressing the crucial need for early and accurate predictions to enhance patient outcomes. Through hyperparameter optimization using GridSearchCV, we significantly improve model performance. Leveraging a range of machine learning (ML) techniques, our approach achieves a high predictive accuracy of 99.07% for random forest, extra trees classifier, logistic regression with L2 penalty, and artificial neural networks (ANN). Through rigorous evaluation, the logistic regression with L2 penalty emerges as the top performer, demonstrating consistent performance. Moreover, integration of Explainable Artificial Intelligence (XAI) techniques, such as Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP), enhances interpretability and reveals insights into model decision-making. By emphasizing an end-to-end model development process, from data collection to deployment, our system enables real-time predictions and informed healthcare decisions. This comprehensive approach underscores the potential of predictive modeling in healthcare to optimize clinical decision-making and improve patient care outcomes.
The evolution of a city is significantly shaped by the design of its urban landscape. The advancement of artificial intelligence has substantially increased convenience for individuals. This research proposes an urban landscape layout model powered by artificial intelligence that automatically generates urban landscape design based on deep learning (URDDL) with two dimensions: emotional tendency and urban landscape appraisal. The input image represents land use and surrounding road conditions, while the output image depicts the selection of the main entrance and the internal spatial function layout. The Pix2Pix model is trained to learn the internal function layout based on varying land use and road conditions. Additionally, a domain-specific dictionary is constructed using an existing semantic resource vocabulary, where positive and negative sentiment words are compared with their corresponding sentiment values, focusing on categories such as Stimulate, Sense, and Action. Experimental results indicate that the absolute average error of the URDDL model is 91.31%, with a maximum error of 96.87%. The degree of fit is highly appropriate for evaluating the emotional prediction of urban landscapes. The findings demonstrate that the URDDL model outperforms traditional design methods regarding generated results, suggesting its potential for future applications in automated landscape design.
This document is a concise summary of many key results in single-agent reinforcement learning (RL). The intended audience are those who already have some familiarity with RL and are looking to review, reference and/or remind themselves of important ideas in the field.
We have designed three search methods for producing the task trees for the provided goal nodes using the Functional Object-Oriented Network. This paper details the strategy, the procedure, and the outcomes.
We investigate the limitations of random trials when the cause of interest is confounded with the effect by formalizing a counterfactual policy-space where the agent's natural predilection is input to a soft-intervention.
"After reading two different crime stories, an artificial intelligence concludes that in both stories the police has found the murderer just by random." -- To what extend and under which assumptions this is a description of a realistic scenario?
This paper investigates the application of consensus clustering and meta-clustering to the set of all possible partitions of a data set. We show that when using a "complement" of Rand Index as a measure of cluster similarity, the total-separation partition, putting each element in a separate set, is chosen.