Information Fusion Research Papers - Academia.edu (original) (raw)

2025, Information Fusion

We present a real-time algorithm for simultaneous localization and local mapping (local SLAM) with detection and tracking of moving objects (DATMO) in dynamic outdoor environments from a moving vehicle equipped with a laser scanner, short range radars and odometry. To correct the vehicle odometry we introduce a new fast implementation of incremental scan matching method that can work reliably in dynamic outdoor environments. After obtaining a good vehicle localization, the map surrounding of the vehicle is updated incrementally and moving objects are detected without a priori knowledge of the targets. Detected moving objects are finally tracked by a Multiple Hypothesis Tracker (MHT) coupled with an adaptive Interacting Multiple Model (IMM) filter. The experimental results on datasets collected from different scenarios such as: urban streets, country roads and highways demonstrate the efficiency of the proposed algorithm.

2025, arXiv (Cornell University)

2D echocardiography is the most common imaging modality for cardiovascular diseases. The portability and relatively low-cost nature of Ultrasound (US) enable the US devices needed for performing echocardiography to be made widely available. However, acquiring and interpreting cardiac US images is operator dependent, limiting its use to only places where experts are present. Recently, Deep Learning (DL) has been used in 2D echocardiography for automated view classification, and structure and function assessment. Although these recent works show promise in developing computer-guided acquisition and automated interpretation of echocardiograms, most of these methods do not model and estimate uncertainty which can be important when testing on data coming from a distribution further away from that of the training data. Uncertainty estimates can be beneficial both during the image acquisition phase (by providing real-time feedback to the operator on acquired image's quality), and during automated measurement and interpretation. The performance of uncertainty models and quantification metric may depend on the prediction task and the models being compared. Hence, to gain insight of uncertainty modelling for left ventricular segmentation from US images, we compare three ensembling based uncertainty models quantified using four different metrics (one newly proposed) on state-of-the-art baseline networks using two publicly available echocardiogram datasets. We further demonstrate how uncertainty estimation can be used to automatically reject poor quality images and improve state-of-the-art segmentation results.

2025, Sensors

Traditionally, classifiers are trained to predict patterns within a feature space. The image classification system presented here trains classifiers to predict patterns within a vector space by combining the dissimilarity spaces generated by a large set of Siamese Neural Networks (SNNs). A set of centroids from the patterns in the training data sets is calculated with supervised k-means clustering. The centroids are used to generate the dissimilarity space via the Siamese networks. The vector space descriptors are extracted by projecting patterns onto the similarity spaces, and SVMs classify an image by its dissimilarity vector. The versatility of the proposed approach in image classification is demonstrated by evaluating the system on different types of images across two domains: two medical data sets and two animal audio data sets with vocalizations represented as images (spectrograms). Results show that the proposed system’s performance competes competitively against the best-per...

2025

The classifier system proposed in this work combines the dissimilarity spaces produced by a set of Siamese neural networks (SNNs) designed using 4 different backbones, with different clustering techniques for training SVMs for automated animal audio classification. The system is evaluated on two animal audio datasets: one for cat and another for bird vocalizations. Different clustering methods reduce the spectrograms in the dataset to a set of centroids that generate (in both a supervised and unsupervised fashion) the dissimilarity space through the Siamese networks. In addition to feeding the SNNs with spectrograms, additional experiments process the spectrograms using the Heterogeneous Auto-Similarities of Characteristics. Once the similarity spaces are computed, a vector space representation of each pattern is generated that is then trained on a Support Vector Machine (SVM) to classify a spectrogram by its dissimilarity vector. Results demonstrate that the proposed approach perfo...

2025, IEEE Journal of Biomedical and Health Informatics

2025

Abstract. Most data mining algorithms assume static behavior of the incoming data. In the real world, the situation is different and most continuously collected data streams are generated by dynamic processes, which may change over time, in some cases even drastically. The change in the underlying concept, also known as concept drift, causes the data mining model generated from past examples to become less accurate and relevant for classifying the current data. Most online learning algorithms deal with concept drift by generating a new model every time a concept drift is detected. On one hand, this solution ensures accurate and relevant models at all times, thus implying an increase in the classification accuracy. On the other hand, this approach suffers from a major drawback, which is the high computational cost of generating new models. The problem is getting worse when a concept drift is detected more frequently and, hence, a compromise in terms of computational effort and accura...

2025

Maintaining systems of military plans is critical for military effectiveness, but is also challenging. Plans will become obsolete as the world diverges from the assumptions on which they rest. If too many ad hoc changes are made to intermeshed plans, the ensemble may no longer lead to well-synchronized and coordinated operations, resulting in the system of plans becoming itself incoherent. We describe in what follows an Adaptive Planning process that we are developing on behalf of the Air Force Research Laboratory (Rome) with the goal of addressing problems of these sorts through cyclical collaborative plan review and maintenance. The interactions of world state, blue force status and associated plans are too complex for manual adaptive processes, and computer-aided plan review and maintenance is thus indispensable. We argue that appropriate semantic technology can 1) provide richer representation of plan-related data and semantics, 2) allow for flexible, non-disruptive, agile, scalable, and coordinated changes in plans, and 3) support more intelligent analytical querying of plan-related data.

2025, 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

2025, Neural Information Processing

Learning from data streams in the presence of concept drifts has become an important application area. When the environment changes, it is necessary to rely on on-line learning with the capability to forget outdated information. Ensemble methods have been among the most successful approaches because they do not need hard-coded and difficult to obtain prior knowledge about the changes in the environment. However, the management of the committee of experts which ultimately controls how past data is forgotten has not been thoroughly investigated so far. This paper shows the importance of the forgetting strategy by comparing several approaches. The results lead us to propose a new ensemble method which compares favorably with the well-known CDC system based on the classical "replace the worst experts" forgetting strategy.

2025, International Journal of Intelligent Systems

The discrete Sugeno integral is an aggregation function particularly suited to aggregation of ordinal inputs. It accounts for inputs interactions, such as redundancy and complementarity, and its learning from empirical data is a challenging optimisation problem. The methods of ordinal regression involve an expensive objective function, whose complexity is quadratic in the number of data. We formulate ordinal regression using a much less expensive objective computed in linear time by the pool-adjacent-violators algorithm. We investigate the learning problem numerically and show the superiority of the new algorithm.

2025, Knowledge Based Systems

This paper examines the marginal contribution representation of fuzzy measures, used to construct fuzzy measure from empirical data through an optimization process. We show that the number of variables can be drastically reduced, and the constraints simplified by using an alternative representation. This technique makes optimizing fitting criteria more efficient numerically, and allows one to tackle learning problems with higher number of correlated decision criteria.

2025, International Seminar on Research of Information Technology and Intelligent Systems (ISRITI), Yogyakarta, Indonesia

Parkinson's disease (PD) is a progressive neurodegenerative disorder, affects motor function and is often challenging to diagnose due to the complex interplay of clinical features. This study integrates a comparative framework integrating hybrid Convolutional Neural Networks (PCNN) and graph-based models (GCN, GAT) to enhance Parkinson's disease (PD) diagnosis using structured medical data. PD, a progressive neurodegenerative disorder affecting motor function, poses diagnostic challenges due to complex clinical feature interactions. The PCNN employs 1D convolutions to capture local feature patterns, while GCN and GAT model intricate interdependencies between clinical variables by representing the dataset as a graph. Notably, GAT's attention mechanism dynamically prioritizes important features, improving interpretability and diagnostic precision. Through hyperparameter optimization with Optuna and addressing class imbalance using SMOTE, our approach achieved a peak accuracy of 97.44%, surpassing traditional methods. The comparative analysis reveals that while PCNN excels in classification accuracy, GAT's attention-based feature selection offers superior interpretability. This makes it a valuable tool for more precise Parkinson's disease detection in clinical applications. The integration of these models provides a comprehensive framework for PD diagnosis, leveraging both local and global feature extraction techniques. This study represents a significant advancement in applying advanced machine learning to neurodegenerative disease diagnostics, offering improved early detection and personalized treatment potential for Parkinson's disease.

2025, IFIP Advances in Information and Communication Technology

Every day the global media system produces an abundance of news stories, all containing many references to people. An important task is to automatically generate reliable lists of people by analysing news content. We describe a system that leverages large amounts of data for this purpose. Lack of structure in this data gives rise to a large number of ways to refer to any particular person. Entity matching attempts to connect references that refer to the same person, usually employing some measure of similarity between references. We use information from multiple sources in order to produce a set of similarity measures with differing strengths and weaknesses. We show how their combination can improve precision without decreasing recall.

2025

Predictive business process monitoring (PBPM) provides a set of techniques to perform different predic tion tasks in running business processes, such as the next activity, the process outcome, or the remaining time. Nowadays, deeplearningbased techniques provide more accurate predictive models. However, the explainability of these models has long been neglected. The predictive quality is essential for PBPM based decision support systems, but also its explainability for human stakeholders needs to be considered. Explainable artificial intelligence (XAI) describes different approaches to make machinelearningbased techniques explainable. To examine the current state of explainable PBPM techniques, we perform a structured and descriptive literature review. We identify explainable PBPM techniques of the domain and classify them along with different XAIrelated concepts: prediction purpose, intrinsically interpretable or posthoc, evaluation objective, and evaluation method. Based o...

2025, 2012 15th International Conference on Information Fusion

Architectural guidance for the development of fusion and resource management systems exists, yet the automated management of their policies remains nascent in the Joint Directors of Laboratories (JDL) model. We present an architecture to assist design of fusion and Command and Control (C2) systems grounded in an understanding of human intent and Artificial Intelligence (AI) literature. The policy automation blueprint provides design guidance by decomposing the problem into levels which align to JDL model levels, and eliciting a process for managing change.

2025, 2012 15th International Conference on Information Fusion

Traditionally the domain of humans, Command and Control (C2) increasingly necessitates the use of automated decision aids and automated decision makers to assist in managing the complexity and dynamics of modern military operations. We propose a blueprint based on the US Joint Directors of Laboratories (JDL) model, cognitive psychology and agent research literature. This blueprint provides design guidance for fusion, resource management and automation policy through decomposition into levels and processes with commensurate levels of human-interface, and the development of foundational language.

2025, Neurocomputing

Recent trends in human-computer interaction (HCI) show a development towards cognitive technical systems (CTS) to provide natural and efficient operating principles. To do so, a CTS has to rely on data from multiple sensors which must be processed and combined by fusion algorithms. Furthermore, additional sources of knowledge have to be integrated, to put the observations made into the correct context. Research in this field often focuses on optimizing the performance of the individual algorithms, rather than reflecting the requirements of CTS. This paper presents the information fusion principles in CTS architectures we developed for Companion Technologies. Combination of information generally goes along with the level of abstractness, time granularity and robustness, such that large CTS architectures must perform fusion gradually on different levelsstarting from sensor-based recognitions to highly abstract logical inferences. In our CTS application we sectioned information fusion approaches into three categories: perception-level fusion, knowledge-based fusion and application-level fusion. For each category, we introduce examples of characteristic algorithms. In addition, we provide a detailed protocol on the implementation performed in order to study the interplay of the developed algorithms.

2025

The goal of this work is to introduce an architecture to automatically detect the amount of stress in the speech signal close to real time. For this an experimental setup to record speech rich in vocabulary and containing different stress levels is presented. Additionally, an experiment explaining the labeling process with a thorough analysis of the labeled data is presented. Fifteen subjects were asked to play an air controller simulation that gradually induced more stress by becoming more difficult to control. During this game the subjects were asked to answer questions, which were then labeled by a different set of subjects in order to receive a subjective target value for each of the answers. A recurrent neural network was used to measure the amount of stress contained in the utterances after training. The neural network estimated the amount of stress at a frequency of 25 Hz and outperformed the human baseline.

2025, Energy-efficient

This paper will examine the need of accurate temperature control in the cooling system of the injection moulding machine to uphold product quality and enhance operational efficiency. The model's inclusion of essential elements such as thermal resistance and cooling efficiency enables a thorough examination of the system dynamics. A comprehensive review of the control and observations was conducted to identify any possible shortcomings in the state estimate and control systems. During the simulation, the temperature was reliably kept within the preferred 24-26 °C range, the advanced MPC algorithm that effectively minimised temperature fluctuations while boosting cooling efficiency. Results show how effective model predictive control is in managing multivariable systems with built-in limitations, providing a significant improvement compared to traditional control methods. The research being conducted not only offers suggestions for boosting production and improving product quality but also strengthens factory management practices. Future research will aim to refine MPC strategies to boost the effectiveness of real-time execution and adjust to evolving operating conditions.

2025, School of Electrical …

2025, CEUR Workshop Proceedings (CEUR-WS.org)

Sentimental Analysis is a process of computing and categorizing the expressed opinions of people about certain event, subject or product as positive, negative or neutral. The major objective of Sentimental Analysis is to help data-driven decisions using insights from replies in social media, surveys and product reviews. Sentiment Analysis can be done with words, sentences, documents, features or aspects, concepts, phrases, links, clauses and implications. Recently, there has been a lot of attention on sentiment analysis especially from researchers in the fields of text mining and natural language processing. But due to extreme absence of annotated datasets which are used to train models in various domains, the accuracy of sentiment analysis has been hindered. Many types of research have been done to confront the challenge and enhance sentiment analysis classification. Sentiment analysis is important as it helps in identifying the emotional and attitude states of people. Positive or negative feelings of people can be expressed in different ways. This research article talks about, in subtle terms, the different ways to deal with sentiment analysis mostly in Machine Learning, Lexicon-based, Hybrid and Ontology-based approaches. This research article gives point by point perspective of the distinctive applications and challenges of Sentiment Analysis.

2025, The Pleks-Krea Theory: A New Understanding of Reality

We present a novel theoretical framework that proposes a fundamental unification of QM phenomena and gravitational efects through
a single entity (Krea) manifesting via projections through a pre-geometric
structure (Pleks). This framework naturally explains the perfect identity of elementary particles, quantum entanglement, and vacuum fluuctuations while making specific, testable predictions about particle behavior in high-energy collisions and near black holes. We present 47
fundamental theorems describing the mathematical structure of this
framework and derive experimental predictions that could validate or
invalidate the theory. The framework suggests new interpretations
of existing experimental data from particle accelerators and quantum
entanglement studies, while proposing new experiments to test its predictions.

2025

Vicomtech-IK4, Spain [agarcia, sgaines, mtlinaza]@vicomtech.org Sentiment analysis has been extensively investigated during the last years mainly for English language. Currently, existing approaches can be split into two main groups: methods based on the combination of lexical resources and Natural Language Processing (NLP) techniques; and machine learning approaches. This paper introduces the use of lexical databases for Sentiment Analysis of user reviews in Spanish for the accommodation and food and beverage sectors. A global sentiment score has been calculate based on the negative and positive words which appear in the review and using the mentioned lexicon database. The algorithm has been tested with short users online reviews acquired from TripAdvisor.

2025, Information Fusion

We propose a novel approach for credit card fraud detection, which combines evidences from current as well as past behavior. The fraud detection system (FDS) consists of four components, namely, rule-based filter, Dempster-Shafer adder, transaction history database and Bayesian learner. In the rule-based component, we determine the suspicion level of each incoming transaction based on the extent of its deviation from good pattern. Dempster-Shafer's theory is used to combine multiple such evidences and an initial belief is computed. The transaction is classified as normal, abnormal or suspicious depending on this initial belief. Once a transaction is found to be suspicious, belief is further strengthened or weakened according to its similarity with fraudulent or genuine transaction history using Bayesian learning. Extensive simulation with stochastic models shows that fusion of different evidences has a very high positive impact on the performance of a credit card fraud detection system as compared to other methods.

2025, Proceedings of the International Computer …

This paper presents work on changepoint detection in musical audio signals, focusing on the case where there are note changes with low associated energy variation. Several methods are described and results of the best are presented.

2025, IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 3, pp. 3886-3889

DUE to the powerful ability of learning hierarchical features,
deep neural networks (DNNs) have achieved great
success in many intelligent perception systems with image,
point cloud, or multi-modal data. To a great degree, the
success of DNNs stems from properly fusing the hierarchical
features, and the representative fusion schemes include dense
connection, residual learning, top-down feature pyramid, and
attention-based feature weighting. However, there is a large
room for developing more effective feature fusion to improve
the performance of DNNs so that the machine perception can
approach or exceed human perception.
This Special Issue focuses on investigating problems and
phenomena of existing feature fusion schemes, and providing
more new ideas, solutions, and insights for effective feature
fusion in DNNs for image, point cloud data, multi-modal data.

2025, IEEE Internet of Things Journal

This work investigates Distributed Detection (DD) in Wireless Sensor Networks (WSNs) utilizing channel-aware binary-decision fusion over a shared flat-fading channel. A reconfigurable metasurface, positioned in the near-field of a limited number of receive antennas, is integrated to enable a holographic Decision Fusion (DF) system. This approach minimizes the need for multiple RF chains while leveraging the benefits of a large array. The optimal fusion rule for a fixed metasurface configuration is derived, alongside two suboptimal joint fusion rule and metasurface design strategies. These suboptimal approaches strike a balance between reduced complexity and lower system knowledge requirements, making them practical alternatives. The design objective focuses on effectively conveying the information regarding the phenomenon of interest to the FC while promoting energy-efficient data analytics aligned with the Internet of Things (IoT) paradigm. Simulation results underscore the viability of holographic DF, demonstrating its advantages even with suboptimal designs and highlighting the significant energy-efficiency gains achieved by the proposed system.

2025

In this study the phoneme labels derived from a phoneme recogniser are used for phonetic speaker recognition. The time-dependencies among phonemes are modelled by using hidden Markov models (HMMs) for the speaker models. Experiments are done using firstorder and second-order HMMs and various smoothing techniques are examined to address the problem of data scarcity. The use of word labels for lexical speaker recognition is also investigated. Single word frequencies are counted and the use of various word selections as feature sets are investigated. During April 2004, the University of Stellenbosch, in collaboration with Spescom DataVoice, participated in an international speaker verification competition presented by the National Institute of Standards and Technology (NIST). The University of Stellenbosch submitted phonetic and lexical (non-acoustic) speaker recognition systems and a fused system (the primary system) that fuses the acoustic system of Spescom DataVoice with the non-acoustic systems of the University of Stellenbosch. The results were evaluated by means of a cost model. Based on the cost model, the primary system obtained second and third position in the two categories that were submitted. Oorsig Hierdie projek maak gebruik van foneem-etikette wat geklassifiseer word deur 'n foneemherkenner en daarna gebruik word vir fonetiese sprekerherkenning. Die tyd-afhanklikhede tussen foneme word gemodelleer deur gebruik te maak van verskuilde Markov modelle (HMMs) as sprekermodelle. Daar word geëksperimenteer met eerste-orde en tweede-orde HMMs en verskeie vergladdingstegnieke word ondersoek om dataskaarsheid aan te spreek. Die gebruik van woord-etikette vir sprekerherkenning word ook ondersoek. Enkelwoordfrekwensies word getel en daar word geëksperimenteer met verskeie woordseleksies as kenmerke vir sprekerherkenning. Gedurende April 2004 het die Universiteit van Stellenbosch in samewerking met Spescom DataVoice deelgeneem aan 'n internasionale sprekerverifikasie kompetisie wat deur die National Institute of Standards and Technology (NIST) aangebied is. Die Universiteit van Stellenbosch het ingeskryf vir 'n fonetiese en 'n woordgebaseerde (nie-akoestiese) sprekerherkenningstelsel, asook 'n saamgesmelte stelsel wat as primêre stelsel dien. Die saamgesmelte stelsel is 'n kombinasie van Spescom DataVoice se akoestiese stelsel en die twee nie-akoestiese stelsels van die Universiteit van Stellenbosch. Die resultate is geëvalueer deur gebruik te maak van 'n koste-model. Op grond van die koste-model het die primêre stelsel tweede en derde plek behaal in die twee kategorieë waaraan deelgeneem is. I would like to thank the following people for their help during the course of this study: -Special thanks to my promotor, Prof. J.A. du Preez, for his aid, guidance, enthusiasm and support. -Andre du Toit for his design of a phoneme recogniser during the 2002 speaker recognition evaluation held by the National Institute of Standards and Technology (NIST). -All the people involved in the NIST 2004 speaker recognition evaluation : Niko Brümmer, Herman Engelbrecht and Francois Cilliers. -Special thanks to my mother, Cynthia du Toit, for all the time she spent to edit this report. Thank you also to Emli-Mari Nel and my promotor, Prof. J.A. du Preez, for proofreading this report. -Thank you to my family and friends, and especially to Ludwig Schwardt, for supporting me during the time of my tribulation.

2025

2025, 2006 IEEE International Symposium on Geoscience and Remote Sensing

Classification of hyperspectral data with high spatial resolution from urban areas is discussed. A previously proposed approach is based on using several principal components from the hyperspectral data to build morphological profiles. These profiles are used all together in one extended morphological profile, which is then classified by a neural network. A shortcoming of the approach is that it is primarily designed for classification of structures and it does not fully utilize the spectral information in the data. An extension is proposed in this paper in order to overcome the problems with the extended morphological profile approach. The proposed method is based on applying data fusion on the original data and the morphological information, after feature extraction. The proposed approach is tested in experiments on two different high resolution remote sensing data sets from urban areas.

2025

This paper describes an approach for political tendency identification of Twitter users. We define some metrics that take into account the polarity of the political entities in the tweets of each user. To obtain this polarities we present the sentiment analysis system developed. The evaluation was performed on the general corpus developed at TASS2013 workshop for Spanish. To our knowledge, the results obtained for the sentiment analysis task and the political tendency identification task are the best results published until now using this data set.

2025, SCIENTIA SINICA Informationis

尤肖虎等: 5G 移动通信发展趋势与若干关键技术移动互联网的蓬勃发展是 5G 移动通信的主要驱动力. 移动互联网将是未来各种新兴业务的基础性业务平台, 现有固定互联网的各种业务将越来越多地通过无线方式提供给用户, 云计算及后台服务的广泛应用将对 5G 移动通信系统提出更高的传输质量与系统容量要求. 5G 移动通信系统的主要发展目标将是与其他无线移动通信技术密切衔接, 为移动互联网的快速发展提供无所不在的基础性业务能力. 按照目前业界的初步估计, 包括 5G 在内的未来无线移动网络业务能力的提升将在 3 个维度上同时进行: 1) 通过引入新的无线传输技术将资源利用率在 4G 的基础上提高 10 倍以上; 2) 通过引入新的体系结构 (如超密集小区结构等) 和更加深度的智能化能力将整个系统的吞吐率提高 25 倍左右; 3) 进一步挖掘新的频率资源 (如高频段、毫米波与可见光等), 使未来无线移动通信的频率资源扩展 4 倍左右. 当前信息技术发展正处于新的变革时期, 5G 技术发展呈现出新的如下特点. 2) 与传统的移动通信系统理念不同, 5G 系统研究将不仅仅把点到点的物理层传输与信道编译码等经典技术作为核心目标, 而是从更为广泛的多点、多用户、多天线、多小区协作组网作为突破的重点, 力求在体系构架上寻求系统性能的大幅度提高. 3) 室内移动通信业务已占据应用的主导地位, 5G 室内无线覆盖性能及业务支撑能力将作为系统优先设计目标, 从而改变传统移动通信系统 "以大范围覆盖为主、兼顾室内" 的设计理念. 4) 高频段频谱资源将更多地应用于 5G 移动通信系统, 但由于受到高频段无线电波穿透能力的限制, 无线与有线的融合、光载无线组网等技术将被更为普遍地应用. 5) 可 "软" 配置的 5G 无线网络将成为未来的重要研究方向, 运营商可根据业务流量的动态变化实时调整网络资源, 有效地降低网络运营的成本和能源的消耗.

2025

This paper describes a security platform as a complex system of holonic communities, that are hierarchically organized, but selfreconfigurable when some of them are detached or cannot otherwise operate. Furthermore, every possible subset of holons may work autonomously, while maintaining self-conscience of its own mission, action lines and goals. Each holonic unit, either elementary or composite, retains some capabilities for sensing (perception), transmissive apparatus (communication), computational processes (elaboration), authentication/authorization (information security), support for data exchange (visualization & interaction), actuators (mission), ambient representation (geometric reasoning), knowledge representation (logic reasoning), situation representation and forecasting (simulation), intelligent feedback (command & control). The higher the organizational level of the holonic unit, the more complex and sophisticated each of its characteristic features.

2025

Multivariate measurements of human brain white matter (WM) with diffusion MRI (dMRI) provide information about the role of WM in a variety of cognitive functions and in brain health. Statistical models take advantage of the regularities in these data to make inferences about individual differences. For example, dMRI data provide the basis for accurate brain-age models – models that predict the chronological age of participants from WM tissue properties. Deep learning (DL) models are powerful machine learning models, which have been shown to provide benefits in many multivariate analysis settings. We investigated whether DL would provide substantial improvements for brain-age models based on dMRI measurements of WM in a large sample of children and adolescents. We found that some DL models fit the data better than a linear baseline, but the differences are small. In particular, recurrent neural network architectures provide up to ∼6% improvement in accuracy. This suggests that inform...

2025, Benchmarking Explainability: A Systematic Evaluation Framework for Interpretable AI in High‑Stakes Decision‑Making

Artificial intelligence (AI) systems have achieved unprecedented performance in diverse tasks, yet their opaque "black-box" nature poses critical challenges. This dissertation develops a comprehensive framework for interpretability and explainability in AI, integrating perspectives from computer science, philosophy of technology, ethics, and law. We begin by establishing a theoretical foundation for explainable AI (XAI), including formal definitions of interpretability and explainability and a taxonomy of techniques (e.g. intrinsic vs post-hoc methods). We trace the historical evolution of these concepts from early expert systems-where decision logic was transparent-to modern deep learning models whose complexity has revived demand for explanations. We then analyze the ethical implications of black-box models in high-stakes contexts, examining how lack of transparency can lead to issues of fairness, accountability, bias, and trust. Philosophical perspectives, including epistemology and socio-technical systems theory, are employed to interrogate what it means to "understand" machine decisions and how explanations function within broader socio-cultural systems. We review the legal and regulatory landscape, focusing on the European Union's General Data Protection Regulation (GDPR) provisions on "meaningful information" about automated decisions and the emerging EU AI Act's transparency and oversight requirements for high-risk AI systems. A case study on algorithmic risk assessment in criminal justice illustrates the tensions between accuracy and interpretability, as evidenced by debates over proprietary sentencing tools that were found to be biased and non-transparent.

2025, STEAM-AI Magazine

What are the skills sets required to be true AI professional ! If you want to become a true AI expert , someone who creates tools and methods, not just uses what others have built , you need to master foundational knowledge across a... more

What are the skills sets required to be true AI professional !
If you want to become a true AI expert , someone who creates tools and methods, not
just uses what others have built , you need to master foundational knowledge across a
few deep areas.

2025, International Journal of Scientific Research and Management (IJSRM)

Stress has become one of the serious concerns in modern society and affects health, both mentally and physically, influencing productivity and quality of life. Chronic stress has grave health consequences, including cardiovascular diseases, anxiety, depression, and suppression of immune responses. Early detection and management are thus of utmost importance for mitigating its adverse effects. The increased popularity of wearable devices and physiological sensors now allows the detection of stressors using realtime data. The presented research is targeted at the study and comparison of the performance of two leading deep learning architectures: LSTM networks and CNN, based on stress detection according to physiological sensor data. This research is based on a dataset of three important physiological parameters: Heart Rate Variability, Galvanic Skin Response, and Skin Temperature, recorded using wearable devices. All these data are filtered for noise and normalized for quality and consistency. In this paper, LSTM and CNN models have been designed, trained, and tested on the same datasets so that a fair comparison can be performed. Accuracy, precision, recall, F1-score, and computational efficiency were the metrics considered to evaluate the model's effectiveness. These results emphasize the fact that LSTM networks outperform CNNs in terms of extracting temporal dependencies in time-series data, yielding high accuracy, recall, and F1-score. This underlines the fact that LSTMs are much more appropriate in applications such as detection of stress, which requires the deeper understanding of sequential patterns. On the other hand, CNNs do have more computational efficiency and multi-dimensional feature extraction capability, useful in cases where real-time performance is required using limited resources. This comparative analysis not only points out the trade-offs of the two models but also serves as a pointer to the potential hybrid approaches which integrate the strengths of LSTM and CNN architectures. The findings of this study give practical insights to the researchers, developers, and practitioners in the domain of health monitoring and stress detection. It therefore enhances the knowledge to date on the ability and limits of the models in the general domain of the application of machine learning in mental health, hence more accurate, scalable, yet efficient automatic stress detection systems are enabled.

2025, arXiv (Cornell University)

Ensembles of neural networks have been shown to give better performance than single networks, both in terms of predictions and uncertainty estimation. Additionally, ensembles allow the uncertainty to be decomposed into aleatoric (data) and epistemic (model) components, giving a more complete picture of the predictive uncertainty. Ensemble distillation is the process of compressing an ensemble into a single model, often resulting in a leaner model that still outperforms the individual ensemble members. Unfortunately, standard distillation erases the natural uncertainty decomposition of the ensemble. We present a general framework for distilling both regression and classification ensembles in a way that preserves the decomposition. We demonstrate the desired behaviour of our framework and show that its predictive performance is on par with standard distillation.

2025, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

This paper is focused on the sensor and information fusion techniques used by a robotic soccer team. Due to the fact that the sensor information is affected by noise, and taking into account the multi-agent environment, these techniques can significantly improve the accuracy of the robot world model. One of the most important elements of the world model is the robot self-localisation. Here, the team localisation algorithm is presented focusing on the integration of visual and compass information. To improve the ball position and velocity reliability, two different techniques have been developed. A study of the visual sensor noise is presented and, according to this analysis, the resulting noise variation depending on the distance is used to define a Kalman filter for ball position. Moreover, linear regression is used for velocity estimation purposes, both for the ball and the robot. This implementation of linear regression has an adaptive buffer size so that, on hard deviations from the path (detected using the Kalman filter), the regression converges more quickly. A team cooperation method based on sharing of the ball position is presented. Besides the ball, obstacle detection and identification is also an important challenge for cooperation purposes. Detecting the obstacles is ceasing to be enough and identifying which obstacles are team mates and opponents is becoming a need. An approach for this identification is presented, considering the visual information, the known characteristics of the team robots and shared localisation among team members. The same idea of distance dependent noise, studied before, is used to improve this identification. Some of the described work, already implemented before RoboCup2008, improved the team performance, allowing it to achieve the 1st place in the Portuguese robotics open Robótica2008 and in the RoboCup2008 world championship.

2025, Mechatronics

When a team of robots is built with the objective of playing soccer, the coordination and control algorithms must reason, decide and actuate based on the current conditions of the robot and its surroundings. This is where sensor and information fusion techniques appear, providing the means to build an accurate model of the world around the robot, based on its own limited sensor information and the also limited information obtained through communication with the team mates. One of the most important elements of the world model is the robot self-localization, as to be able to decide what to do in an effective way, it must know its position in the field of play. In this paper, the team localization algorithm is presented focusing on the integration of visual and compass information. An important element in a soccer game, perhaps the most important, is the ball. To improve the estimations of the ball position and velocity, two different techniques have been developed. A study of the visual sensor noise is presented and, according to this analysis, the resulting noise variation is used to define the parameters of a Kalman filter for ball position estimation. Moreover, linear regression is used for velocity estimation purposes, both for the ball and the robot. This implementation of linear regression has an adaptive buffer size so that, on hard deviations from the path (detected using the Kalman filter), the regression converges faster. A team cooperation method based on sharing the ball position is presented. Other important data during the soccer game is obstacle data. This is an important challenge for cooperation purposes, allowing the improvement of team strategy with ball covering, dribble corridor estimation, pass lines, among other strategic possibilities. Thus, detecting the obstacles is ceasing to be enough and identifying which obstacles are team mates and opponents is becoming a need. An approach for this identification is presented, considering the visual information, the known characteristics of the team robots and shared localization among team members. The described work was implemented on the CAMBADA team and allowed it to achieve particularly good performances in the last two years, with a 1st and a 3rd place in the world championship RoboCup 2008 and RoboCup 2009 editions, respectively, as well as distinctively achieve 1st place in 2008 and 2009 editions of the Portuguese Robotics Open.

2025, ssrn

Semi-supervised learning (SSL) presents a compelling approach to harness both labelled and unlabelled data, meeting the increasing need for effective machine learning models in circumstances where labelled data is limited. Although much progress has been made, the current state of SSL frameworks still falls short in scalability, noise-robustness, domainadaptability, and interpretability. We present a novel Scalable and Robust Universal Semi-Supervised Learning Framework that combines various key components including dynamic feature transformation, hybrid loss functions, domain adaptation and explainable AI techniques. It outperforms existing models on benchmark datasets covering a wide variety of domains text, image, and graph data in particular. Experimental results demonstrate that the framework can cope with both noisy and sparse datasets, adapt to domain shifts, and provide interpretable predictions through SHAP explain ability and attention visualization. The proposed framework is versatile and efficient in practice, thus, deepening the pathway towards next-generation models for semi-supervised learning.

2025, Proceedings of the Fifth International Conference on Information Fusion. FUSION 2002. (IEEE Cat.No.02EX5997)

In support of the AFOSR program in Information Fusion, the CNS Technology Laboratory at Boston University is developing and applying neural models of image and signal processing, pattern learning and recognition, associative learning dynamics, and 3D visualization, to the domain of Information Fusion for Image Analysis in a geospatial context. Our research is focused by a challenge problem involving the emergence of a crisis in an urban environment, brought on by a terrorist attack or other man-made or natural disaster. We aim to develop methods aiding preparation and monitoring of the battlespace, deriving context from multiple sources of imagery (high-resolution visible and low-resolution hyperspectral) and signals (GMTI from moving vehicles, and ELINT from emitters). This context will serve as a foundation, in conjunction with existing knowledge nets, for exploring neural methods in higherlevel information fusion supporting situation assessment and creation of a common operating picture (COP).

2025, IEEE

We introduce a framework that utilizes dynamic nowledge graphs (DKGs), graph neural networks (GNNs) and natural language processing (NLP) to provide predictive business insights. This integration allows for efficient data extraction, real-time updates, and cross-domain applicability. This seamless integration not only enhances scalability and explainability but also integrates advanced security measures, providing businesses with actionable insights leading to faster and more strategic decision-making. Experimental results show improvements in handling large and dynamic datasets with no loss in accuracy or performance. The framework demonstrates how diverse networks from the fields of retail, healthcare, and finance can leverage generalization capabilities to enable decision makers to derive strategic actions from the available data. Notably, this work highlights the enabling power of DKG for a new paradigm of dynamic real-time intelligent decisionmaking that addresses the intricacies of business process dynamics. Utilizing the power of advanced AI, the framework fills the void between unrefined data and actionable insights, establishing itself as a powerful asset for tackling contemporary business challenges.

2025, Saudi journal of engineering and technology

Over recent years there is a huge growth in the area of wireless networks due to its efficient design. The word sensor is defined as component which handles & monitor different kinds of inputs from different sources such as physical and environmental conditions like pressure, heat, light, sound and vibrations etc. The output produced by the sensor will be electrical in nature and this electrical signal is further applied to various controllers for other functions. A Wireless Sensor Network is basically a form of ad-hoc network which consists of thousands of tiny sensor nodes. These wireless networks are deployed where the wired network implementation is difficult or impossible. These nodes are further distributed over a wide area such as hilly areas, forests, deserts, ocean etc. and these tiny sensors communicate or exchange data with each other by using radio signals. The WSN utilizes various communication protocols like Bluetooth, Wi-Fi, Zigbee and Ultra Wide Band techniques. Every protocol has its own speed and depending upon the distance between them, there are various problems which network can face like battery failure, effective coverage area, and effective energy utilization or node failure. In this paper, the several different methods to build energy efficient network, methods to improve the lifetime of a network, method for detection and correction of node failure and in the last the important applications are discussed.

2025

The Internet of Things (IoT) is an emerging technology that tends to increasingly become part of everyday life. The multiple aspects of IoT, and the increasingly large number of devices, technologies and platforms in the field, led to the development of IoT technology in many areas. The Internet of Things is a technology that will allow the entry into a new economic era for the entire globe. IoT is a concept that defines a world where all the objects (cars, appliances, lighting systems, mobile devices, portable devices etc.) are connected to each other via the Internet. In this paper, there are presented the most important characteristics and major applications of IoT, and the technological challenges that the Internet of Things is facing. IoT is not the result of a single new technology, several complementary technical advances provide capabilities that, together, help to bridge the gap between the virtual world and the physical one. These capabilities support the IoT and the prosp...

2025, Journal of Next-Generation Research 5.0 (JNGR 5.0)Journal of Next-Generation Research 5.0 (JNGR 5.0)

This study is the third part of a research project involving analyzing written documents to assess emotional charges. Those documents have been written following a three-day trading experience. The first part focused on the textual analysis of these documents using several measurement scales. The second part consisted of a comparative analysis of the results using three Artificial Intelligences (ChatGPT, Gemini, and DeepSeek), working through several queries designed to clarify the emotional fields under analysis as well as the general context that led to the documents being written. Our results demonstrated the ability of Artificial Intelligence to identify a general emotional context but revealed difficulties in approaching some emotional nuances. In this third study, we refined the queries addressed to the three Artificial Intelligences, particularly about the emotions to be taken into consideration, making a distinction between positive anticipation and negative anticipation, as well as a distinction between a good surprise and a bad surprise. The emotional typology used prevented us from considering how the two aspects could differ. Finally, the Artificial Intelligences were informed that a reward would be granted to the best-performing portfolio at the end of the experiment. Our results show that DeepSeek gives great importance to this parameter and generates many positive emotions, which neither ChatGPT nor Gemini do. Our results once again demonstrate a consistent ability of Artificial Intelligence to clarify the prevailing general emotional context, but difficulties in identifying consistently and uniformly the emotional nuances associated with the experiment. Concluding these three articles, it seems clear that Artificial Intelligences could be used to begin the process of understanding qualitative data, but it must be complemented by a human dimension in order to capture the emotional nuances that Artificial Intelligences analysis cannot.

2025, The role of random forest in internal audit to enhance financial reporting accuracy

trust and informed corporate decision-making. With the proliferation of complex financial transactions, audit teams face mounting challenges in deciphering voluminous transactional data to safeguard financial reporting quality. Machine learning has the potential to identify signifiers of financial reporting quality. Within the Design Science Methodology framework, we apply the Random Forest Classifier technique to metrics such as the error rate to enhance financial reporting. We find that the Random Forest Classifier identifies that certain parameters are critical to error detection, which enhance account receivable accuracy, lower receivable account control risk. This research advances the argument that technologically-enhanced internal audit procedures can play a pivotal role in ensuring that financial reporting mirrors the economic reality of the company.