AI, Volume 5, Issue 4 (December 2024) – 36 articles

Accurately forecasting power consumption is crucial important for efficient energy management. Machine learning (ML) models are often employed for this purpose. However, tuning their hyperparameters is a complex and time-consuming task. The article presents a novel multi-objective (MO) hybrid evolutionary-based approach, GA-SHADE-MO, for tuning ML models aimed at solving the complex problem of forecasting power consumption. The proposed algorithm simultaneously optimizes both hyperparameters and feature sets across six different ML models, ensuring enhanced accuracy and efficiency. The study focuses on predicting household power consumption at hourly and daily levels. The hybrid MO evolutionary algorithm integrates elements of genetic algorithms and self-adapted differential evolution. By incorporating MO optimization, GA-SHADE-MO balances the trade-offs between model complexity (the number of used features) and prediction accuracy, ensuring robust performance across various forecasting scenarios. Experimental numerical results show the superiority of the proposed method compared to traditional tuning techniques, and random search, showcasing significant improvements in predictive accuracy and computational efficiency. The findings suggest that the proposed GA-SHADE-MO approach offers a powerful tool for optimizing ML models in the context of energy consumption forecasting, with potential applications in other domains requiring precise predictive modeling. The study contributes to the advancement of ML optimization techniques, providing a framework that can be adapted and extended for various predictive analytics tasks. Full article

Machine learning (ML) techniques have permeated various domains, offering intelligent solutions to complex problems. ML has been increasingly explored for applications in active power filters (APFs) due to its potential to enhance harmonic compensation, reference signal generation, filter control optimization, and fault detection and diagnosis. This paper reviews the most recent applications of ML in APFs, highlighting their abilities to adapt to nonlinear load conditions, improve fault detection and classification accuracy, and optimize system performance in real time. However, this paper also highlights several limitations of these methods, such as the high computational complexity, the need for extensive training data, and challenges with real-time deployment in distributed power systems. For example, the marginal improvements in total harmonic distortion (THD) achieved by ML-based methods often do not justify the increased computational overhead compared to traditional control methods. This review then suggests future research directions to overcome these limitations, including lightweight ML models for faster and more efficient control, federated learning for decentralized optimization, and digital twins for real-time system monitoring. While traditional methods remain effective, ML-based solutions have the potential to significantly enhance APF performance in future power systems. Full article

(1) Background: The mutual relationship between evolution and learning is a controversial argument among the artificial intelligence and neuro-evolution communities. After more than three decades, there is still no common agreement on the matter. (2) Methods: In this paper, the author investigates whether combining learning and evolution permits finding better solutions than those discovered by evolution alone. In further detail, the author presents a series of empirical studies that highlight some specific conditions determining the success of such combination. Results are obtained in five qualitatively different domains: (i) the 5-bit parity task, (ii) the double-pole balancing problem, (iii) the Rastrigin, Rosenbrock and Sphere optimization functions, (iv) a robot foraging task and (v) a social foraging problem. Moreover, the first three tasks represent benchmark problems in the field of evolutionary computation. (3) Results and discussion: The outcomes indicate that the effect of learning on evolution depends on the nature of the problem. Specifically, when the problem implies limited or absent agent–environment conditions, learning is beneficial for evolution, especially with the introduction of noise during the learning and selection processes. Conversely, when agents are embodied and actively interact with the environment, learning does not provide advantages, and the addition of noise is detrimental. Finally, the absence of stochasticity in the experienced conditions is paramount for the effectiveness of the combination. Furthermore, the length of the learning process must be fine-tuned based on the considered task. Full article

The widespread use of machine and deep learning algorithms for anomaly detection has created a critical need for robust explanations that can identify the features contributing to anomalies. However, effective evaluation methodologies for anomaly explanations are currently lacking, especially those that compare the explanations against the true underlying causes, or ground truth. This paper aims to address this gap by introducing a rigorous, ground-truth-based framework for evaluating anomaly explanation methods, which enables the assessment of explanation correctness and robustness—key factors for actionable insights in anomaly detection. To achieve this, we present an innovative benchmark dataset of digital circuit truth tables with model-based anomalies, accompanied by local ground truth explanations. These explanations were generated using a novel algorithm designed to accurately identify influential features within each anomaly. Additionally, we propose an evaluation methodology based on correctness and robustness metrics, specifically tailored to quantify the reliability of anomaly explanations. This dataset and evaluation framework are publicly available to facilitate further research and standardize evaluation practices. Our experiments demonstrate the utility of this dataset and methodology by evaluating common model-agnostic explanation methods in an anomaly detection context. The results highlight the importance of ground-truth-based evaluation for reliable and interpretable anomaly explanations, advancing both theory and practical applications in explainable AI. This work establishes a foundation for rigorous, evidence-based assessments of anomaly explanations, fostering greater transparency and trust in AI-driven anomaly detection systems. Full article

As machine learning (ML) transforms industries, the need for efficient model development tools using high-performance computing (HPC) and ensuring interpretability is crucial. This paper presents SIBILA, an AutoML approach designed for HPC environments, focusing on the interpretation of ML models. SIBILA simplifies model development by allowing users to set objectives and preferences before automating the search for optimal ML pipelines. Unlike traditional AutoML frameworks, SIBILA is specifically designed to exploit the computational capabilities of HPC platforms, thereby accelerating the model search and evaluation phases. The emphasis on interpretability is particularly crucial when model transparency is mandated by regulations or desired for stakeholder understanding. SIBILA has been validated in different tasks with public datasets. The results demonstrate that SIBILA consistently produces models with competitive accuracy while significantly reducing computational overhead. This makes it an ideal choice for practitioners seeking efficient and transparent ML solutions on HPC infrastructures. SIBILA is a major advancement in AutoML, addressing the rising demand for explainable ML models on HPC platforms. Its integration of interpretability constraints alongside automated model development processes marks a substantial step forward in bridging the gap between computational efficiency and model transparency in ML applications. The tool is available as a web service at no charge. Full article

Background: This article introduces SC-Phi2, a fine-tuned StarCraft II small language model. Small language models, like Phi2, Gemma, and DistilBERT, are streamlined versions of large language models (LLMs) with fewer parameters that require less computational power and memory to run. Method: To teach Microsoft’s Phi2 model about StarCraft, we create a new SC2 text dataset with information about StarCraft races, roles, and actions and use it to fine-tune Phi-2 with self-supervised learning. We pair this language model with a Vision Transformer (ViT) from the pre-trained BLIP-2 (Bootstrapping Language Image Pre-training) model, fine-tuning it on the StarCraft replay dataset, MSC. This enables us to construct dynamic prompts that include visual game state information. Results: Unlike the large models used in StarCraft LLMs such as GPT-3.5, Phi2 is trained primarily on textbook data and contains little inherent knowledge of StarCraft II beyond what is provided by our training process. By using LoRA (Low-rank Adaptation) and quantization, our model can be trained on a single GPU. We demonstrate that our model performs well at build order prediction, an important StarCraft macromanagement task. Conclusions: Our research on the usage of small models is a step towards reducing the carbon footprint of AI agents. Full article

The irruption of GenAI such as ChatGPT has changed the educational landscape. Therefore, methodological guidelines and more empirical experiences are needed to better understand these tools and know how to use them to their fullest potential. This contribution presents an exploratory and correlational study conducted with 37 computer science students who used ChatGPT as a support tool to learn database administration. The article addresses three questions: The first one explores the degree of use of ChatGPT among computer science students to learn database administration, the second one explores the profile of students who get the most out of tools like ChatGPT to deal with database administration activities, and the third one explores how the utilization of ChatGPT can impact in academic performance. To empirically shed light on these questions the student’s grades and a comprehensive questionnaire were employed as research instruments. The obtained results indicate that traditional learning resources, such as teacher’s explanations and student’s reports, were widely used and correlated positively with student’s grades. The usage and perceived utility of ChatGPT were moderate, but positive correlations between students’ grades and ChatGPT usage were found. Indeed, a significantly higher use of this tool was identified among the group of outstanding students. This indicate that high-performing students are the ones who are using ChatGPT the most. So, a new digital trench could be rising between these students and those with a lower degree of fundamentals and worse prompting skills, who may not take advantage of all the ChatGPT possibilities. Full article

(1) Background: Detecting people and technical objects in various situations, such as natural disasters and warfare, is critical to search and rescue operations and the safety of civilians. A fast and accurate detection of people and equipment can significantly increase the effectiveness of search and rescue missions and provide timely assistance to people. Computer vision and deep learning technologies play a key role in detecting the required objects due to their ability to analyze big volumes of visual data in real-time. (2) Methods: The performance of the neural networks such as You Only Look Once (YOLO) v4-v8, Faster R-CNN, Single Shot MultiBox Detector (SSD), and EfficientDet has been analyzed using COCO2017, SARD, SeaDronesSee, and VisDrone2019 datasets. The main metrics for comparison were mAP, Precision, Recall, F1-Score, and the ability of the neural network to work in real-time. (3) Results: The most important metrics for evaluating the efficiency and performance of models for a given task are accuracy (mAP), F1-Score, and processing speed (FPS). These metrics allow us to evaluate both the accuracy of object recognition and the ability to use the models in real-world environments where high processing speed is important. (4) Conclusion: Although different neural networks perform better on certain types of metrics, YOLO outperforms them on all metrics, showing the best results of mAP-0.88, F1-0.88, and FPS-48, so the focus was on these models. Full article

Background: The Internet of Things (IoT) has improved many aspects that have impacted the industry and the people’s daily lives. To begin with, the IoT allows communication to be made across a wide range of devices, from household appliances to industrial machinery. This connectivity allows for a better integration of the pervasive computing, making devices “smart” and capable of interacting with each other and with the corresponding users in a sublime way. However, the widespread adoption of IoT devices has introduced some security challenges, because these devices usually run in environments that have limited resources. As IoT technology becomes more integrated into critical infrastructure and daily life, the need for stronger security measures will increase. These devices are exposed to a variety of cyber-attacks. This literature review synthesizes the current research of artificial intelligence (AI) technologies to improve IoT security. This review addresses key research questions, including: (1) What are the primary challenges and threats that IoT devices face?; (2) How can AI be used to improve IoT security?; (3) What AI techniques are currently being used for this purpose?; and (4) How does applying AI to IoT security differ from traditional methods? Methods: We included a total of 33 peer-reviewed studies published between 2020 and 2024, specifically in journal and conference papers written in English. Studies irrelevant to the use of AI for IoT security, duplicate studies, and articles without full-text access were excluded. The literature search was conducted using scientific databases, including MDPI, ScienceDirect, IEEE Xplore, and SpringerLink. Results were synthesized through a narrative synthesis approach, with the help of the Parsifal tool to organize and visualize key themes and trends. Results: We focus on the use of machine learning, deep learning, and federated learning, which are used for anomaly detection to identify and mitigate the security threats inherent to these devices. AI-driven technologies offer promising solutions for attack detection and predictive analysis, reducing the need for human intervention more significantly. This review acknowledges limitations such as the rapidly evolving nature of IoT technologies, the early-stage development or proprietary nature of many AI techniques, the variable performance of AI models in real-world applications, and potential biases in the search and selection of articles. The risk of bias in this systematic review is moderate. While the study selection and data collection processes are robust, the reliance on narrative synthesis and the limited exploration of potential biases in the selection process introduce some risk. Transparency in funding and conflict of interest reporting reduces bias in those areas. Discussion: The effectiveness of these AI-based approaches can vary depending on the performance of the model and the computational efficiency. In this article, we provide a comprehensive overview of existing AI models applied to IoT security, including machine learning (ML), deep learning (DL), and hybrid approaches. We also examine their role in enhancing the detection accuracy. Despite all the advances, challenges still remain in terms of data privacy and the scalability of AI solutions in IoT security. Conclusion: This review provides a comprehensive overview of ML applications to enhance IoT security. We also discuss and outline future directions, emphasizing the need for collaboration between interested parties and ongoing innovation to address the evolving threat landscape in IoT security. Full article

Background: Advances in medical image classification have recently benefited from general augmentation techniques. However, these methods often fall short in performance and interpretability. Objective: This paper applies the Unified Model Agnostic Computation (UMAC) framework specifically to the medical domain to demonstrate its utility in this critical area. Methods: UMAC is a model-agnostic methodology designed to develop machine learning approaches that integrate seamlessly with various paradigms, including self-supervised, semi-supervised, and supervised learning. By unifying and standardizing computational models and algorithms, UMAC ensures adaptability across different data types and computational environments while incorporating state-of-the-art methodologies. In this study, we integrate UMAC as a plug-and-play module within convolutional neural networks (CNNs) and Transformer architectures, enabling the generation of high-quality representations even with minimal data. Results: Our experiments across nine diverse 2D medical image datasets show that UMAC consistently outperforms traditional data augmentation methods, achieving a 1.89% improvement in classification accuracy. Conclusions: Additionally, by incorporating explainable AI (XAI) techniques, we enhance model transparency and reliability in decision-making. This study highlights UMAC’s potential as a powerful tool for improving both the performance and interpretability of medical image classification models. Full article

App reviews provide crucial feedback for software maintenance and evolution, but manually extracting useful reviews from vast volumes is time-consuming and challenging. This study investigates the effectiveness of six Naïve Bayes variants for automatically filtering useful app reviews. We evaluated these variants on datasets from five popular apps, comparing their performance in terms of accuracy, precision, recall, F-measure, and processing time. Our results show that Expectation Maximization-Multinomial Naïve Bayes with Laplace smoothing performed best overall, achieving up to 89.2% accuracy and 0.89 F-measure. Complement Naïve Bayes with Laplace smoothing demonstrated particular effectiveness for imbalanced datasets. Generally, incorporating Laplace smoothing and Expectation Maximization improved performance, albeit with increased processing time. This study also examined the impact of data imbalance on classification performance. Our findings suggest that these advanced Naïve Bayes variants hold promise for filtering useful app reviews, especially when dealing with limited labeled data or imbalanced datasets. This research contributes to the body of evidence around app review mining and provides insights for enhancing software maintenance and evolution processes. Full article

The artificial bee colony (ABC) algorithm is a famous swarm intelligence method utilized across various disciplines due to its robustness. However, it exhibits limitations in exploration mechanisms, particularly in high-dimensional or complex landscapes. This article introduces the adaptive exploration artificial bee colony (AEABC), a novel variant that reinspires the ABC algorithm based on real-world phenomena. AEABC incorporates new distance-based parameters and mechanisms to correct the original design, enhancing its robustness. The performance of AEABC was evaluated against 33 state-of-the-art metaheuristics across twenty-five benchmark functions and an engineering application. AEABC consistently outperformed its counterparts, demonstrating superior efficiency and accuracy. In a variable-sized problem (n = 10), the traditional ABC algorithm converged to 3.086 × 10⁶, while AEABC achieved a convergence of 2.0596 × 10⁻²⁵⁵, highlighting its robust performance. By addressing the shortcomings of the traditional ABC algorithm, AEABC significantly advances mathematical optimization, especially in engineering applications. This work underscores the significance of the inspiration of the traditional ABC algorithm in enhancing the capabilities of swarm intelligence. Full article

This study analyzes the nonlinear optical properties exhibited by graphene, focusing on the nonlinear absorption coefficient and the nonlinear refractive index. The evaluation was conducted using the Z-scan technique with a 532 nm wavelength laser at various intensities. The nonlinear optical absorption and the nonlinear optical refractive index were measured. Four machine learning models, including linear regression, decision trees, random forests, and gradient boosting regression, were trained to analyze how the nonlinear optical absorption coefficient varies with variables such as spot radius, maximum energy, and normalized minimum transmission. The models were trained with synthetic data and subsequently validated with experimental data. Decision tree-based models, such as random forests and gradient boosting regression, demonstrated superior performance compared to linear regression, especially in terms of mean squared error. This work provides a detailed assessment of the nonlinear optical properties of graphene and highlights the effectiveness of machine learning methods in this context. Full article

Background: Dynamic multiobjective optimization problems (DMOPs) involve multiple conflicting and time-varying objectives, and dynamic multiobjective algorithms (DMOAs) aim to find Pareto optima that are closer to the real one in the new environment as soon as possible. In particular, the introduction of transfer learning in DMOAs has led to good results in solving DMOPs. However, the selection of valuable historical knowledge and the mitigation of negative transfer remain important problems in existing transfer learning-based DMOAs. Method: A DMOA based on multi-environment knowledge selection and transfer (MST-DMOA) is proposed in this article. First, by clustering historical Pareto optima, some representative solutions that can reflect the main evolutionary information are selected as knowledge of the environment. Second, the similarity between the historical and current environments is evaluated, and then the knowledge of multiple similar environments is selected as valuable historical knowledge to construct the source domain. Third, solutions with high quality in the new environment are obtained to form the target domain, which can better help historical knowledge to adapt to the current environment, thus effectively alleviating negative transfer. Conclusions: We compare the proposed MST-DMOA with five state-of-the-art DMOAs on fourteen benchmark test problems, and the experimental results verify the excellent performance of MST-DMOA in solving DMOPs. Full article

Human action recognition has become crucial in computer vision, with growing applications in surveillance, human–computer interaction, and healthcare. Traditional approaches often use broad feature representations, which may miss subtle variations in timing and movement within action sequences. Our proposed One-to-Many Hierarchical Contrastive Learning (OTM-HC) framework maps the input into multi-layered feature vectors, creating a hierarchical contrast representation that captures various granularities within a human skeleton sequence temporal and spatial domains. Using sequence-to-sequence (Seq2Seq) transformer encoders and downsampling modules, OTM-HC can distinguish between multiple levels of action representations, such as instance, domain, clip, and part levels. Each level contributes significantly to a comprehensive understanding of action representations. The OTM-HC model design is adaptable, ensuring smooth integration with advanced Seq2Seq encoders. We tested the OTM-HC framework across four datasets, demonstrating improved performance over state-of-the-art models. Specifically, OTM-HC achieved improvements of 0.9% and 0.6% on NTU60, 0.4% and 0.7% on NTU120, and 0.7% and 0.3% on PKU-MMD I and II, respectively, surpassing previous leading approaches across these datasets. These results showcase the robustness and adaptability of our model for various skeleton-based action recognition tasks. Full article

The global community is awaiting the advent of a self-driving vehicle that is safe, reliable, and capable of navigating a diverse range of road conditions and terrains. This requires a lot of research, study, and optimization. Thus, this work focused on implementing, training, and optimizing a convolutional neural network (CNN) model, aiming to predict the steering angle during driving (one of the main issues). The considered dataset comprises images collected inside a car-driving simulator and further processed for augmentation and removal of unimportant details. In addition, an innovative data-balancing process was previously performed. A CNN model was trained with the dataset, conducting a comparison between several different standard optimizers. Moreover, evolutionary optimization was applied to optimize the model’s weights as well as the optimizers themselves. Several experiments were performed considering different approaches of genetic algorithms (GAs) along with other optimizers from the state of the art. The obtained results demonstrate that the GA is an effective optimization tool for this problem. Full article

Recent research has drawn attention to the ambiguity surrounding the definition and learnability of Out-of-Distribution recognition. Although the original problem remains unsolved, the term “Out-of-Model Scope” detection offers a clearer perspective. The ability to detect Out-of-Model Scope inputs is particularly beneficial in safety-critical applications such as autonomous driving or medicine. By detecting Out-of-Model Scope situations, the system’s robustness is enhanced and it is prevented from operating in unknown and unsafe scenarios. In this paper, we propose a novel approach for Out-of-Model Scope detection that integrates three sources of information: (1) the original input, (2) its latent feature representation extracted by an encoder, and (3) a synthesized version of the input generated from its latent representation. We demonstrate the effectiveness of combining original and synthetically generated inputs to defend against adversarial attacks in the computer vision domain. Our method, TRust Your GENerator (TRYGEN), achieves results comparable to those of other state-of-the-art methods and allows any encoder to be integrated into our pipeline in a plug-and-train fashion. Through our experiments, we evaluate which combinations of the encoder’s features are most effective for discovering Out-of-Model Scope samples and highlight the importance of a compact feature space for training the generator. Full article

Objective: This paper describes the development of hybrid artificial intelligence strategies for drone navigation. Methods: The navigation module combines a deep learning model with a rule-based engine depending on the agent state. The deep learning model has been trained using reinforcement learning. The rule-based engine uses expert knowledge to deal with specific situations. The navigation module incorporates several strategies to explain the drone decision based on its observation space, and different mechanisms for including human decisions in the navigation process. Finally, this paper proposes an evaluation methodology based on defining several scenarios and analyzing the performance of the different strategies according to metrics adapted to each scenario. Results: Two main navigation problems have been studied. For the first scenario (reaching known targets), it has been possible to obtain a 90% task completion rate, reducing significantly the number of collisions thanks to the rule-based engine. For the second scenario, it has been possible to reduce 20% of the time required to locate all the targets using the reinforcement learning model. Conclusions: Reinforcement learning is a very good strategy to learn policies for drone navigation, but in critical situations, it is necessary to complement it with a rule-based module to increase task success rate. Full article

A machine learning method was applied to solve an inverse airfoil design problem. A conditional VAE-WGAN-gp model, which couples the conditional variational autoencoder (VAE) and Wasserstein generative adversarial network with gradient penalty (WGAN-gp), is proposed for an airfoil generation method, and then, it is compared with the WGAN-gp and VAE models. The VAEGAN model couples the VAE and GAN models, which enables feature extraction in the GAN models. In airfoil generation tasks, to generate airfoil shapes that satisfy lift coefficient requirements, it is known that VAE outperforms WGAN-gp with respect to the accuracy of the reproduction of the lift coefficient, whereas GAN outperforms VAE with respect to the smoothness and variations of generated shapes. In this study, VAE-WGAN-gp demonstrated a good performance in all three aspects. Latent distribution was also studied to compare the feature extraction ability of the proposed method. Full article

Machine learning (ML) has transformed the financial industry by enabling advanced applications such as credit scoring, fraud detection, and market forecasting. At the core of this transformation is deep learning (DL), a subset of ML that is robust in processing and analyzing complex and large datasets. This paper provides a comprehensive overview of key deep learning models, including Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), Deep Belief Networks (DBNs), Transformers, Generative Adversarial Networks (GANs), and Deep Reinforcement Learning (Deep RL). Beyond summarizing their mathematical foundations and learning processes, this study offers new insights into how these models are applied in real-world financial contexts, highlighting their specific advantages and limitations in tasks such as algorithmic trading, risk management, and portfolio optimization. It also examines recent advances and emerging trends in the financial industry alongside critical challenges such as data quality, model interpretability, and computational complexity. These insights can guide future research directions toward developing more efficient, robust, and explainable financial models that address the evolving needs of the financial sector. Full article

Autosomal dominant polycystic kidney disease (ADPKD) is the predominant hereditary factor leading to end-stage renal disease (ESRD) worldwide, affecting individuals across all races with a prevalence of 1 in 400 to 1 in 1000. The disease presents significant challenges in management, particularly with limited options for slowing cyst progression, as well as the use of tolvaptan being restricted to high-risk patients due to potential liver injury. However, determining high-risk status typically requires magnetic resonance imaging (MRI) to calculate total kidney volume (TKV), a time-consuming process demanding specialized expertise. Motivated by these challenges, this study proposes alternative methods for high-risk categorization that do not rely on TKV data. Utilizing historical patient data, we aim to predict rapid kidney enlargement in ADPKD patients to support clinical decision-making. We applied seven machine learning algorithms—Random Forest, Logistic Regression, Support Vector Machine (SVM), Light Gradient Boosting Machine (LightGBM), Gradient Boosting Tree, XGBoost, and Deep Neural Network (DNN)—to data from the Polycystic Kidney Disease Outcomes Consortium (PKDOC) database. The XGBoost model, combined with the Synthetic Minority Oversampling Technique (SMOTE), yielded the best performance. We also leveraged explainable artificial intelligence (XAI) techniques, specifically Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP), to visualize and clarify the model’s predictions. Furthermore, we generated text summaries to enhance interpretability. To evaluate the effectiveness of our approach, we proposed new metrics to assess explainability and conducted a survey with 27 doctors to compare models with and without XAI techniques. The results indicated that incorporating XAI and textual summaries significantly improved expert explainability and increased confidence in the model’s ability to support treatment decisions for ADPKD patients. Full article

Innovative wood inspection technology is crucial in various industries, especially for determining wood quality by counting rings in each stave, a key factor in wine barrel production. (1) Background: Traditionally, human inspectors visually evaluate staves, compensating for natural variations and characteristics like dirt and saw-induced aberrations. These variations pose significant challenges for automatic inspection systems. Several techniques using classical image processing and deep learning have been developed to detect tree-ring boundaries, but they often struggle with woods exhibiting heterogeneity and texture irregularities. (2) Methods: This study proposes a hybrid approach combining classical computer vision techniques for preprocessing with deep learning algorithms for classification, designed for continuous automated processing. To enhance performance and accuracy, we employ a data augmentation strategy using cropping techniques to address intra-class variability in individual staves. (3) Results: Our approach significantly improves accuracy and reliability in classifying wood with irregular textures and heterogeneity. The use of explainable AI and model calibration offers a deeper understanding of the model’s decision-making process, ensuring robustness and transparency, and setting confidence thresholds for outputs. (4) Conclusions: The proposed system enhances the performance of automatic wood inspection technologies, providing a robust solution for industries requiring precise wood quality assessment, particularly in wine barrel production. Full article

Various Artificial Intelligence (AI) techniques in water resource management highlight the current methodologies’ strengths and limitations in forecasting, optimization, and control. We identify a gap in integrating these diverse approaches for enhanced water prediction and management. We critically analyze the existing literature on artificial neural networks (ANNs), deep learning (DL), long short-term memory (LSTM) networks, machine learning (ML) models such as supervised learning (SL) and unsupervised learning (UL), and random forest (RF). In response, we propose a novel framework that synergizes these techniques into a unified, multi-layered model and incorporates a digital twin and a multi-modal transformer approach. This integration aims to leverage the collective advantages of each method while overcoming individual constraints, significantly enhancing prediction accuracy and operational efficiency. This paper sets the foundation for an innovative digital twin-integrated solution, focusing on reviewing past works as a precursor to a detailed exposition of our proposed model in a subsequent publication. This advanced approach promises to redefine accuracy in water demand forecasting and contribute significantly to global sustainability and efficiency in water use. Full article

This paper presents a new approach in the field of probability-informed machine learning (ML). It implies improving the results of ML algorithms and neural networks (NNs) by using probability models as a source of additional features in situations where it is impossible to increase the training datasets for various reasons. We introduce connected mixture components as a source of additional information that can be extracted from a mathematical model. These components are formed using probability mixture models and a special algorithm for merging parameters in the sliding window mode. This approach has been proven effective when applied to real-world time series data for short- and medium-term forecasting. In all cases, the models informed by the connected mixture components showed better results than those that did not use them, although different informed models may be effective for various datasets. The fundamental novelty of the research lies both in a new mathematical approach to informing ML models and in the demonstrated increase in forecasting accuracy in various applications. For geophysical spatiotemporal data, the decrease in Root Mean Square Error (RMSE) was up to $27.7\%$, and the reduction in Mean Absolute Percentage Error (MAPE) was up to $45.7\%$ compared with ML models without probability informing. The best metrics values were obtained by an informed ensemble architecture that fuses the results of a Long Short-Term Memory (LSTM) network and a transformer. The Mean Squared Error (MSE) for the electricity transformer oil temperature from the ETDataset had improved by up to $10.0\%$ compared with vanilla methods. The best MSE value was obtained by informed random forest. The introduced probability-informed approach allows us to outperform the results of both transformer NN architectures and classical statistical and machine learning methods. Full article

Background: Nonsurgical treatment of uncomplicated appendicitis is a reasonable option in many cases despite the sparsity of robust, easy access, externally validated, and multimodally informed clinical decision support systems (CDSSs). Developed by OpenAI, the Generative Pre-trained Transformer 3.5 model (GPT-3) may provide enhanced decision support for surgeons in less certain appendicitis cases or those posing a higher risk for (relative) operative contra-indications. Our objective was to determine whether GPT-3.5, when provided high-throughput clinical, laboratory, and radiological text-based information, will come to clinical decisions similar to those of a machine learning model and a board-certified surgeon (reference standard) in decision-making for appendectomy versus conservative treatment. Methods: In this cohort study, we randomly collected patients presenting at the emergency department (ED) of two German hospitals (GFO, Troisdorf, and University Hospital Cologne) with right abdominal pain between October 2022 and October 2023. Statistical analysis was performed using R, version 3.6.2, on RStudio, version 2023.03.0 + 386. Overall agreement between the GPT-3.5 output and the reference standard was assessed by means of inter-observer kappa values as well as accuracy, sensitivity, specificity, and positive and negative predictive values with the “Caret” and “irr” packages. Statistical significance was defined as p < 0.05. Results: There was agreement between the surgeon’s decision and GPT-3.5 in 102 of 113 cases, and all cases where the surgeon decided upon conservative treatment were correctly classified by GPT-3.5. The estimated model training accuracy was 83.3% (95% CI: 74.0, 90.4), while the validation accuracy for the model was 87.0% (95% CI: 66.4, 97.2). This is in comparison to the GPT-3.5 accuracy of 90.3% (95% CI: 83.2, 95.0), which did not perform significantly better in comparison to the machine learning model (p = 0.21). Conclusions: This study, the first study of the “intended use” of GPT-3.5 for surgical treatment to our knowledge, comparing surgical decision-making versus an algorithm found a high degree of agreement between board-certified surgeons and GPT-3.5 for surgical decision-making in patients presenting to the emergency department with lower abdominal pain. Full article

Recent significant advances in the healthcare industry due to artificial intelligence (AI) and machine learning (ML) have been shown to revolutionize healthcare delivery by improving efficiency, accuracy, and patient outcomes. However, these technologies can face significant challenges and ethical considerations. This systematic review aimed to gather and synthesize the current knowledge on the impact of AI and ML adoption in healthcare delivery, with its associated challenges and opportunities. This study adhered to the PRISMA guidelines. Articles from 2014 to 2024 were selected from various databases using specific keywords. Eligible studies were included after rigorous screening and quality assessment using checklist tools. Themes were identified through data analysis and thematic analysis. From 4981 articles screened, a data synthesis of nine eligible studies revealed themes, including productivity enhancement, improved patient care through decision support and precision medicine, legal and policy challenges, technological considerations, organizational and managerial aspects, ethical concerns, data challenges, and socioeconomic implications. There exist significant opportunities, as well as substantial challenges and ethical concerns, associated with integrating AI and ML into healthcare delivery. Implementation strategies must be carefully designed, considering technical, ethical, and social factors. Full article

This paper proposes causal economic machine learning (CEML) as a research agenda that utilizes causal machine learning (CML), built on causal economics (CE) decision theory. Causal economics is better suited for use in machine learning optimization than expected utility theory (EUT) and behavioral economics (BE) based on its central feature of causal coupling (CC), which models decisions as requiring upfront costs, some certain and some uncertain, in anticipation of future uncertain benefits that are linked by causation. This multi-period causal process, incorporating certainty and uncertainty, replaces the single-period lottery outcomes augmented with intertemporal discounting used in EUT and BE, providing a more realistic framework for AI machine learning modeling and real-world application. It is mathematically demonstrated that EUT and BE are constrained versions of CE. With the growing interest in natural experiments in statistics and causal machine learning (CML) across many fields, such as healthcare, economics, and business, there is a large potential opportunity to run AI models on CE foundations and compare results to models based on traditional decision-making models that focus only on rationality, bounded to various degrees. To be most effective, machine learning must mirror human reasoning as closely as possible, an alignment established through CEML, which represents an evolution to truly “human AI”. This paper maps out how the non-linear optimization required for the CEML structural response functions can be accomplished through Sequential Least Squares Programming (SLSQP) and applied to data sets through the S-Learner CML meta-algorithm. Upon this foundation, the next phase of research is to apply CEML to appropriate data sets in various areas of practice where causality and accurate modeling of human behavior are vital, such as precision healthcare, economic policy, and marketing. Full article

Visual thinking leverages spatial mechanisms in animals for navigation and reasoning. Therefore, given the challenge of abstract mathematics and logic, spatial reasoning-based teaching strategies can be highly effective. Our previous research verified that innovative box-and-ball coloring activities help teach elementary school students complex notions like quantifiers, logical connectors, and dynamic systems. However, given the richness of the activities, correction is slow, error-prone, and demands high attention and cognitive load from the teacher. Moreover, feedback to the teacher should be immediate. Thus, we propose to provide the teacher with real-time help with LLMs. We explored various prompting techniques with and without context—Zero-Shot, Few-Shot, Chain of Thought, Visualization of Thought, Self-Consistency, logicLM, and emotional —to test GPT-4o’s visual, logical, and correction capabilities. We obtained that Visualization of Thought and Self-Consistency techniques enabled GPT-4o to correctly evaluate 90% of the logical–spatial problems that we tested. Additionally, we propose a novel prompt combining some of these techniques that achieved 100% accuracy on a testing sample, excelling in spatial problems and enhancing logical reasoning. Full article

The robots that entered the manufacturing sector in the second and third Industrial Revolutions (IR2 and IR3) were designed for carrying out predefined routines without physical interaction with humans. In contrast, IR4* robots (i.e., robots since IR4 and beyond) are supposed to interact with humans in a cooperative way for enhancing flexibility, autonomy, and adaptability, thus dramatically improving productivity. However, human–robot cooperation implies cognitive capabilities that the cooperative robots (CoBots) in the market do not have. The common wisdom is that such a cognitive lack can be filled in a straightforward way by integrating well-established ICT technologies with new AI technologies. This short paper expresses the view that this approach is not promising and suggests a different one based on artificial cognition rather than artificial intelligence, founded on concepts of embodied cognition, developmental robotics, and social robotics. We suggest giving these IR4* robots designed according to such principles the name CoCoBots. The paper also addresses the ethical problems that can be raised in cases of critical emergencies. In normal operating conditions, CoCoBots and human partners, starting from individual evaluations, will routinely develop joint decisions on the course of action to be taken through mutual understanding and explanation. In case a joint decision cannot be reached and/or in the limited case that an emergency is detected and declared by top security levels, we suggest that the ultimate decision-making power, with the associated responsibility, should rest on the human side, at the different levels of the organized structure. Full article

In the context of digitization and automation, extracting relevant information from business documents remains a significant challenge. It is typical to rely on machine-learning techniques to automate the process, reduce manual labor, and minimize errors. This work introduces a new model for extracting key values from electricity invoices, including customer data, bill breakdown, electricity consumption, or marketer data. We evaluate several machine learning techniques, such as Naive Bayes, Logistic Regression, Random Forests, or Support Vector Machines. Our approach relies on a bag-of-words strategy and custom-designed features tailored for electricity data. We validate our method on the IDSEM dataset, which includes 75,000 electricity invoices with eighty-six fields. The model converts PDF invoices into text and processes each word separately using a context of eleven words. The results of our experiments indicate that Support Vector Machines and Random Forests perform exceptionally well in capturing numerous values with high precision. The study also explores the advantages of our custom features and evaluates the performance of unseen documents. The precision obtained with Support Vector Machines is 91.86% on average, peaking at 98.47% for one document template. These results demonstrate the effectiveness of our method in accurately extracting key values from invoices. Full article