Big Data Cogn. Comput., Volume 9, Issue 3 (March 2025) – 24 articles

Adversarial attacks threaten the reliability of machine learning models in critical applications like autonomous vehicles and defense systems. As object detectors become more robust with models like YOLOv8, developing effective adversarial methodologies is increasingly challenging. We present Truck Adversarial Camouflage Optimization (TACO), a novel framework that generates adversarial camouflage patterns on 3D vehicle models to deceive state-of-the-art object detectors. Adopting Unreal Engine 5, TACO integrates differentiable rendering with a Photorealistic Rendering Network to optimize adversarial textures targeted at YOLOv8. To ensure the generated textures are both effective in deceiving detectors and visually plausible, we introduce the Convolutional Smooth Loss function, a generalized smooth loss function. Experimental evaluations demonstrate that TACO significantly degrades YOLOv8’s detection performance, achieving an AP@0.5 of 0.0099 on unseen test data. Furthermore, these adversarial patterns exhibit strong transferability to other object detection models such as Faster R-CNN and earlier YOLO versions. Full article

Accurately forecasting CO₂ emissions in the transportation sector is essential for developing effective mitigation strategies. This study uses an annually spanning dataset from 1993 to 2022 to evaluate the predictive performance of three methods: NAR, NARX, and GA-T2FIS. Among these, NARX-VK, which incorporates vehicle kilometers (VK) and economic variables, demonstrated the highest predictive accuracy, achieving a MAPE of 2.2%, MAE of 1621.449 × 10³ tons, and RMSE of 1853.799 × 10³ tons. This performance surpasses that of NARX-RG, which relies on registered vehicle data and achieved a MAPE of 3.7%. While GA-T2FIS exhibited slightly lower accuracy than NARX-VK, it demonstrated robust performance in handling uncertainties and nonlinear relationships, achieving a MAPE of 2.6%. Sensitivity analysis indicated that changes in VK significantly influence CO₂ emissions. The Green Transition Scenario, assuming a 10% reduction in VK, led to a 4.4% decrease in peak CO₂ emissions and a 4.1% reduction in total emissions. Conversely, the High Growth Scenario, modeling a 10% increase in VK, resulted in a 7.2% rise in peak emissions and a 4.1% increase in total emissions. Full article

Climate change is one of the most pressing global threats, endangering the sustainability of the planet and quality of life, whilst urban mobility significantly contributes to exacerbating its effects. Recently, policies aimed at mitigating these effects have been implemented, emphasizing the promotion of sustainable travel culture. Prior research has indicated that both environmental awareness and regulatory efforts could encourage the shift towards greener mobility; however, factors that affect young people’s travel behavior remain understudied. This study examined whether and how climate change impacts travel behavior, particularly among Generation Z in Greece. A comprehensive online survey was conducted, from 31 March to 8 April 2024, within a Greek academic community, yielding 904 responses from Generation Z individuals. The design of the survey was informed by an adaptation of Triandis’ Theory of Interpersonal Behavior. The study also incorporated a comparative analysis using data from the UK’s National Travel Attitudes Survey (NTAS), offering insights from a different cultural and socio-economic context. Blending an Exploratory Factor Analysis and latent variable ordered probit and logit models, the key determinants of the willingness to reduce car use and self-reported reduction in car use in response to climate change were identified. The results indicate that emotional factors, social roles, and norms, along with socio-demographic characteristics, current behaviors, and local environmental concerns, significantly influence car-related travel choices among Generation Z. For instance, concerns about local air quality are consistently correlated with a higher likelihood of having already reduced car use due to climate change and a higher willingness to reduce car travel in the future. The NTAS data reveal that flexibility in travel habits and social norms are critical determinants of the willingness to reduce car usage. The findings of the study highlight the key role of policy interventions, such as the implementation of Low-Emission Zones, leveraging social media for environmental campaigns, and enhancing infrastructure for active travel and public transport to foster broader cultural shifts towards sustainable travel behavior among Generation Z. Full article

Threads is a new social network that was launched by Meta in July 2023 and conceived as a direct alternative to X. It is a unique case study in the social network landscape, as it is content-based like X, but has an Instagram-based growth model, which makes it significantly different from X. As it was launched recently, studies on Threads are still scarce. One of the most common investigations in social networks regards power users (also called influencers, lead users, influential users, etc.), i.e., those users who can significantly influence information dissemination, user behavior, and ultimately the current dynamics and future development of a social network. In this paper, we want to contribute to the knowledge of Threads by showing that there are indeed power users in this social network and then attempt to understand the main features that characterize them. The definition of power users that we adopt here is novel and leverages the four classical centrality measures of Social Network Analysis. This ensures that our study of power users can benefit from the enormous knowledge on centrality measures that has accumulated in the literature over the years. In order to conduct our analysis, we had to build a Threads dataset, as none existed in the literature that contained the information necessary for our studies. Once we built such a dataset, we decided to make it open and thus available to all researchers who want to perform analyses on Threads. This dataset, the new definition of power users, and the characterization of Threads power users are the main contributions of this paper. Full article

The explicit kernel transformation of input data vectors to their distributed high-dimensional representations has recently been receiving increasing attention in the field of hyperdimensional computing (HDC). The main argument is that such representations endow simpler last-leg classification models, often referred to as HDC classifiers. HDC models have obvious advantages over resource-intensive deep learning models for use cases requiring fast, energy-efficient computations both for model training and deploying. Recent approaches to training HDC classifiers have primarily focused on various methods for selecting individual learning rates for incorrectly classified samples. In contrast to these methods, we propose an alternative strategy where the decision to learn is based on a margin applied to the classifier scores. This approach ensures that even correctly classified samples within the specified margin are utilized in training the model. This leads to improved test performances while maintaining a basic learning rule with a fixed (unit) learning rate. We propose and empirically evaluate two such strategies, incorporating either an additive or multiplicative margin, on the standard subset of the UCI collection, consisting of 121 datasets. Our approach demonstrates superior mean accuracy compared to other HDC classifiers with iterative error-correcting training. Full article

We present an automated journal recommendation pipeline designed to evaluate the performance of five Sentence Transformer models—all-mpnet-base-v2 (Mpnet), all-MiniLM-L6-v2 (Minilm-l6), all-MiniLM-L12-v2 (Minilm-l12), multi-qa-distilbert-cos-v1 (Multi-qa-distilbert), and all-distilroberta-v1 (roberta)—for recommending journals aligned with a manuscript’s thematic scope. The pipeline extracts domain-relevant keywords from a manuscript via KeyBERT, retrieves potentially related articles from PubMed, and encodes both the test manuscript and retrieved articles into high-dimensional embeddings. By computing cosine similarity, it ranks relevant journals based on thematic overlap. Evaluations on 50 test articles highlight mpnet’s strong performance (mean similarity score 0.71 ± 0.04), albeit with higher computational demands. Minilm-l12 and minilm-l6 offer comparable precision at lower cost, while multi-qa-distilbert and roberta yield broader recommendations better suited to interdisciplinary research. These findings underscore key trade-offs among embedding models and demonstrate how they can provide interpretable, data-driven insights to guide journal selection across varied research contexts. Full article

This paper introduces PK-Judge, a novel neural network watermarking framework designed to enhance the intellectual property (IP) protection by incorporating an asymmetric cryptograp hic approach in the verification process. Inspired by the paradigm shift from HTTP to HTTPS in enhancing web security, this work integrates public key infrastructure (PKI) principles to establish a secure and verifiable watermarking system. Unlike symmetric approaches, PK-Judge employs a public key infrastructure (PKI) to decouple ownership validation from the extraction process, significantly increasing its resilience against adversarial attacks. Additionally, it incorporates a robust challenge-response mechanism to mitigate replay attacks and leverages error correction codes (ECC) to achieve an Effective Bit Error Rate (EBER) of zero, ensuring watermark integrity even under conditions such as fine-tuning, pruning, and overwriting. Furthermore, PK-Judge introduces a new requirement based on the principle of separation of privilege, setting a foundation for secure and scalable watermarking mechanisms in machine learning. By addressing these critical challenges, PK-Judge advances the state-of-the-art in neural network IP protection and integrity, paving the way for trust-based AI technologies that prioritize security and verifiability. Full article

Hierarchical classification, which organizes items into structured categories and subcategories, has emerged as a powerful solution for handling large and complex datasets. However, traditional flat classification approaches often overlook the hierarchical dependencies between classes, leading to suboptimal predictions and limited interpretability. This paper addresses these challenges by proposing a novel integration of tree-based models with hierarchical-aware split criteria through adjusted entropy calculations. The proposed method calculates entropy at multiple hierarchical levels, ensuring that the model respects the taxonomic structure during training. This approach aligns statistical optimization with class semantic relationships, enabling more accurate and coherent predictions. Experiments conducted on real-world datasets structured according to the GS1 Global Product Classification (GPC) system demonstrate the effectiveness of our method. The proposed model was applied using tree-based ensemble methods combined with the newly developed hierarchy-aware metric Penalized Information Gain (PIG). PIG was implemented with level-wise entropy adjustments, assigning greater weight to higher hierarchical levels to maintain the taxonomic structure. The model was trained and evaluated on two real-world datasets based on the GS1 Global Product Classification (GPC) system. The final dataset included approximately 30,000 product descriptions spanning four hierarchical levels. An 80-20 train–test split was used, with model hyperparameters optimized through 5-fold cross-validation and Bayesian search. The experimental results showed a 12.7% improvement in classification accuracy at the lowest hierarchy level compared to traditional flat classification methods, with significant gains in datasets featuring highly imbalanced class distributions and deep hierarchies. The proposed approach also increased the F1 score by 12.6%. Despite these promising results, challenges remain in scaling the model for very large datasets and handling classes with limited training samples. Future research will focus on integrating neural networks with hierarchy-aware metrics, enhancing data augmentation to address class imbalance, and developing real-time classification systems for practical use in industries such as retail, logistics, and healthcare. Full article

A key challenge in utilizing the expected goals on target (xGOT) metric is the limited public access to detailed football event and positional data, alongside other advanced metrics. This study aims to develop an xGOT model to evaluate goalkeeper (GK) performance based on the probability of successful actions, considering not only the outcomes (saves or goals conceded) but also the difficulty of each shot faced. Formal definitions were established for the following: (i) the initial distance between the ball and the GK at the moment of the shot, (ii) the distance between the ball and the GK over time post-shot, and (iii) the distance between the GK’s initial position and the goal, with respect to the y-coordinate. An xGOT model incorporating geometric parameters was designed to optimize performance based on the ball position, trajectory, and GK positioning. The model was tested using shots on target from the 2022 FIFA World Cup. Statistical evaluation using k-fold cross-validation yielded an AUC-ROC score of 0.67 and an 85% accuracy, confirming the model’s ability to differentiate successful GK performances. This approach enables a more precise evaluation of GK decision-making by analyzing a representative dataset of shots to estimate the probability of success. Full article

Inter-state cyberattacks are increasingly becoming a major hidden threat to national security and global order. However, current prediction models are often constrained by single-source data due to insufficient consideration of complex influencing factors, resulting in limitations in understanding and predicting cyberattacks. To address this issue, we comprehensively consider multiple data sources including cyberattacks, bilateral interactions, armed conflicts, international trade, and national attributes, and propose an interpretable multimodal data fusion framework for predicting cyberattacks among countries. On one hand, we design a dynamic multi-view graph neural network model incorporating temporal interaction attention and multi-view attention, which effectively captures time-varying dynamic features and the importance of node representations from various modalities. Our proposed model exhibits greater performance in comparison to many cutting-edge models, achieving an F1 score of 0.838. On the other hand, our interpretability analysis reveals unique characteristics of national cyberattack behavior. For example, countries with different income levels show varying preferences for data sources, reflecting their different strategic focuses in cyberspace. This unveils the factors and regional differences that affect cyberattack prediction, enhancing the transparency and credibility of the proposed model. Full article

As generative AI (GenAI) technologies proliferate, ensuring trust and transparency in digital ecosystems becomes increasingly critical, particularly within democratic frameworks. This article examines decentralized Web3 mechanisms—blockchain, decentralized autonomous organizations (DAOs), and data cooperatives—as foundational tools for enhancing trust in GenAI. These mechanisms are analyzed within the framework of the EU’s AI Act and the Draghi Report, focusing on their potential to support content authenticity, community-driven verification, and data sovereignty. Based on a systematic policy analysis, this article proposes a multi-layered framework to mitigate the risks of AI-generated misinformation. Specifically, as a result of this analysis, it identifies and evaluates seven detection techniques of trust stemming from the action research conducted in the Horizon Europe Lighthouse project called ENFIELD: (i) federated learning for decentralized AI detection, (ii) blockchain-based provenance tracking, (iii) zero-knowledge proofs for content authentication, (iv) DAOs for crowdsourced verification, (v) AI-powered digital watermarking, (vi) explainable AI (XAI) for content detection, and (vii) privacy-preserving machine learning (PPML). By leveraging these approaches, the framework strengthens AI governance through peer-to-peer (P2P) structures while addressing the socio-political challenges of AI-driven misinformation. Ultimately, this research contributes to the development of resilient democratic systems in an era of increasing technopolitical polarization. Full article

Generative AI (GenAI) models are designed to produce realistic and natural data, such as images, audio, or written text. Due to their high computational and memory demands, these models traditionally run on powerful remote compute servers. However, there is growing interest in deploying GenAI models at the edge, on resource-constrained embedded devices. Since 2018, the TinyML community has proved that running fixed topology AI models on edge devices offers several benefits, including independence from internet connectivity, low-latency processing, and enhanced privacy. Nevertheless, deploying resource-consuming GenAI models on embedded devices is challenging since the latter have limited computational, memory, and energy resources. This review paper aims to evaluate the progresses made to date in the field of Edge GenAI, an emerging area of research within the broader domain of EdgeAI which focuses on bringing GenAI on edge devices. Papers released between 2022 and 2024 that address the design and deployment of GenAI models on embedded devices are identified and described. Additionally, their approaches and results are compared. This manuscript contributes to understand the ongoing transition from TinyML to Edge GenAI and provides valuable insights to the AI research community on this emerging, impactful, and quite under-explored field. Full article

This paper explores advancements in real-time talking-head generation, focusing on overcoming challenges in Audio Feature Extraction (AFE), which often introduces latency and limits responsiveness in real-time applications. To address these issues, we propose and implement a fully integrated system that replaces conventional AFE models with OpenAI’s Whisper, leveraging its encoder to optimize processing and improve overall system efficiency. Our evaluation of two open-source real-time models across three different datasets shows that Whisper not only accelerates processing but also improves specific aspects of rendering quality, resulting in more realistic and responsive talking-head interactions. Although interviewer training systems are considered a potential application, the primary contribution of this work is the improvement of the technical foundations necessary for creating responsive AI avatars. These advancements enable more immersive interactions and expand the scope of AI-driven applications, including educational tools and simulated training environments. Full article

The detection of synthetic speech has become a pressing challenge due to the potential societal risks posed by synthetic speech technologies. Existing methods primarily focus on either the time or frequency domain of speech, limiting their ability to generalize to new and diverse speech synthesis algorithms. In this work, we present a novel and scientifically grounded approach, the Dual-domain Fusion Network (DDFNet), which synergistically integrates features from both the time and frequency domains to capture complementary information. The architecture consists of two specialized single-domain feature extraction networks, each optimized for the unique characteristics of its respective domain, and a feature fusion network that effectively combines these features at a deep level. Moreover, we incorporate multi-task learning to simultaneously capture rich, multi-faceted representations, further enhancing the model’s generalization capability. Extensive experiments on the ASVspoof 2019 Logical Access corpus and ASVspoof 2021 tracks demonstrate that DDFNet achieves strong performance, maintaining competitive results despite the challenges posed by channel changes and compression coding, highlighting its robust generalization ability. Full article

We present an automatic approach for generating learning problems for teaching introductory programming in different programming languages. The current implementation allows input and output in the three most popular programming languages for teaching introductory programming courses: C++, Java, and Python. The generator stores learning problems using the “meaning tree”, a language-independent representation of a syntax tree. During this study, we generated a bank of 1,428,899 learning problems focused on the order of expression evaluation. They were generated in about 16 h. The learning problems were classified for further use with the used concepts, possible domain-rule violations, and required skills; they covered a wide range of difficulties and topics. The problems were validated by automatically solving them in an intelligent tutoring system that recorded the actual skills used and violations made. The generated problems were favorably assessed by 10 experts: teachers and teaching assistants in introductory programming courses. They noted that the problems are ready for use without further manual improvement and that the classification system is flexible enough to receive problems with desirable properties. The proposed approach combines the advantages of different state-of-the-art methods. It combines the diversity of learning problems generated by restricted randomization and large language models with full correctness and a natural look of template-based problems, which makes it a good fit for large-scale learning problem generation. Full article

This paper critically examines the expanding body of literature on ChatGPT, a transformative AI tool with widespread global adoption. By categorising research into six key themes—sustainability, health, education, work, social media, and energy—it explores ChatGPT’s versatility, benefits, and challenges. The findings highlight its potential to enhance productivity, streamline workflows, and improve access to knowledge while also revealing critical limitations, including high energy consumption, informational inaccuracies, and ethical concerns. The paper underscores the need for robust regulatory frameworks, sustainable AI practices, and interdisciplinary collaboration to optimise benefits while mitigating risks. Future research should focus on improving ChatGPT’s reliability, inclusivity, and environmental sustainability to ensure its responsible integration across diverse sectors. Full article

What makes a wine exceptional enough to score a perfect 10 from experts? This study explores a data-driven approach to identify the ideal physicochemical composition for wines that could achieve this highest possible rating. Using a dataset of 11 measurable attributes, including alcohol, sulfates, residual sugar, density, and citric acid, for wines rated up to a maximum quality score of 8 by expert tasters, we sought to predict compositions that might enhance wine quality beyond current observations. Our methodology applies a second-degree polynomial ridge regression model, optimized through an exhaustive evaluation of feature combinations. Furthermore, we propose a specific chemical and physical composition of wine that our model predicts could achieve a quality score of 10 from experts. While further validation with winemakers and industry experts is necessary, this study aims to contribute a practical tool for guiding quality exploration and advancing predictive modeling applications in food and beverage sciences. Full article

Cognitive computing encompasses computing tools and methods that simulate and mimic the process of human thinking, without human supervision. Deep neural network architectures, natural language processing, big data tools, and self-learning tools based on pattern recognition have been widely deployed to solve highly complex problems. Business intelligence enhances collaboration among different organizational departments with data-driven conversations and provides an organization with meaningful data interpretation for making strategic decisions on time. Since the introduction of ChatGPT in November 2022, the tremendous impacts of using Large Language Models have been rippling through cognitive computing, business intelligence, and their applications in accounting, finance, and management. Unlike other recent reviews in related areas, this review focuses precisely on the cognitive computing perspective, with frontier applications in accounting, finance, and management. Some current limitations and future directions of cognitive computing are also discussed. Full article

An image–text retrieval method that integrates intramodal fine-grained local semantic information and intermodal global semantic information is proposed to address the weak fine-grained discrimination capabilities for the semantic features located between image and text modalities in cross-modal retrieval tasks. First, the original features of images and texts are extracted, and a graph attention network is employed for region relationship reasoning to obtain relation-enhanced local features. Then, an attention mechanism is used for different semantically interacting samples within the same modality, enabling comprehensive intramodal relationship learning and producing semantically enhanced image and text embeddings. Finally, a triplet loss function is used to train the entire model, and it is enhanced with an angular constraint. Through extensive comparative experiments conducted on the Flickr30K and MS-COCO benchmark datasets, the effectiveness and superiority of the proposed method were verified. It outperformed the current method by 6.4% relatively for image retrieval and 1.3% relatively for caption retrieval on MS-COCO (Recall@1 using the 1K test set). Full article

Hand impairment affects millions of people. There are multiple factors that cause deficits, varying from physical injuries to neurological disorders. Upper-limb patients face significant difficulties in daily life. Rehabilitation aims at supporting them to regain functionality and increasing their independence and quality of life. Assessment is key to therapy, as it offers an evaluation of the condition of patients, leading to suitable treatments. Unfortunately, rehabilitation relies on clinical resources, making it expensive and time-consuming. Digital technology can provide solutions that make treatments more flexible and affordable. With the use of computer vision, we created an online platform that includes several exercises and serious games, based on movements and gestures performed in real-world treatments. Difficulty levels vary, and therapists can monitor these procedures remotely, while performance can be stored and tracked over time, identifying improvement. There is no need for any special equipment, as the platform can be accessed like a common website and all its applications require only a simple computer camera and stable Internet connection. In this article, we present our research approach, we analyze the development of the platform, and we provide a brief demonstration of its use in practice. Furthermore, we address some technical challenges and we share the results derived from preliminary test phases, concluding by outlining future plans. Full article

The complex and specialized terminology of financial language in Portuguese-speaking markets create significant challenges for natural language processing (NLP) applications, which must capture nuanced linguistic and contextual information to support accurate analysis and decision-making. This paper presents DeB3RTa, a transformer-based model specifically developed through a mixed-domain pretraining strategy that combines extensive corpora from finance, politics, business management, and accounting to enable a nuanced understanding of financial language. DeB3RTa was evaluated against prominent models—including BERTimbau, XLM-RoBERTa, SEC-BERT, BusinessBERT, and GPT-based variants—and consistently achieved significant gains across key financial NLP benchmarks. To maximize adaptability and accuracy, DeB3RTa integrates advanced fine-tuning techniques such as layer reinitialization, mixout regularization, stochastic weight averaging, and layer-wise learning rate decay, which together enhance its performance across varied and high-stakes NLP tasks. These findings underscore the efficacy of mixed-domain pretraining in building high-performance language models for specialized applications. With its robust performance in complex analytical and classification tasks, DeB3RTa offers a powerful tool for advancing NLP in the financial sector and supporting nuanced language processing needs in Portuguese-speaking contexts. Full article

Large language models (LLMs) have demonstrated remarkable capabilities in text generation, which also raise numerous concerns about their potential misuse, especially in educational exercises and academic writing. Accurately identifying and tracing the origins of LLM-generated content is crucial for accountability and transparency, ensuring the responsible use of LLMs in educational and academic environments. Previous methods utilize binary classifiers to discriminate whether a piece of text was written by a human or generated by a specific LLM or employ multi-class classifiers to trace the source LLM from a fixed set. These methods, however, are restricted to one or several pre-specified LLMs and cannot generalize to new LLMs, which are continually emerging. This study formulates source LLM tracing in a class-incremental learning (CIL) fashion, where new LLMs continually emerge, and a model incrementally learns to identify new LLMs without forgetting old ones. A training-free continual learning method is further devised for the task, the idea of which is to continually extract prototypes for emerging LLMs, using a frozen encoder, and then to perform origin tracing via prototype matching after a delicate decorrelation process. For evaluation, two datasets are constructed, one in English and one in Chinese. These datasets simulate a scenario where six LLMs emerge over time and are used to generate student essays, and an LLM detector has to incrementally expand its recognition scope as new LLMs appear. Experimental results show that the proposed method achieves an average accuracy of 97.04% on the English dataset and 91.23% on the Chinese dataset. These results validate the feasibility of continual origin tracing of LLM-generated text and verify its effectiveness in detecting cheating in student coursework. Full article

There is a rapid growth in mental disorders, thus leading to a pressing demand for more sophisticated diagnosis techniques. Clinical assessments and symptomatic analyses for traditional diagnostics suffer from subjectivity, delayed diagnosis, and specificity deficiencies. Therefore, this study developed the Psychological Disorders Machine Learning Genomic (PDMLG) model as an amalgamation of genetic algorithms and machine learning techniques in a predictive analysis model using genomic data samples. The two central components of the PDMLG model include the Genomic Fusion Model, which uses ensemble learning techniques like Random Forest, Gradient Boosting, and Neural Networks, and Deep Learning Model of Convolutional and Recurrent Neural Networks in processing genomic sequence data samples. The model enhanced the disease classification and early detection where the model achieved improvement in precision, recall, and specificity by 3.5% to 9.4% compared to the baseline methods Near Neighbor-Boundary Enlargement (NNBE), Collaborative Mmatrix Factorization based on Correntropy (LDCMFC), and Microsatellite Instability (MSI). The area under the curve of this model is up to 94.95%, which reflects the model’s robust performance on a variety of diseases like Schizophrenia, Bipolar Disorders, and Alzheimer’s. In addition, the PDMLG model can indicate important genetic markers, and this is vital for understanding the genetic basis of psychological conditions that may be diagnosed early and treatment plans prepared in advance for this process. This is a step forward in personalized medicine, which could revolutionize clinical practice in mental disorders diagnostics. This would not be substituted for the established psychological or doctor evaluations. However, it was considered a complementary tool auxiliary for the professional know-how and gives data-related insights that the professional should corroborate for this. Full article