<?xml version="1.0" encoding="utf-8"?>
    <rss version="2.0">
      <channel xmlns:content="http://purl.org/rss/1.0/modules/content/">
        <title>Frontiers in Artificial Intelligence | Machine Learning and Artificial Intelligence section | New and Recent Articles</title>
        <link>https://www.frontiersin.org/journals/artificial-intelligence/sections/machine-learning-and-artificial-intelligence</link>
        <description>RSS Feed for Machine Learning and Artificial Intelligence section in the Frontiers in Artificial Intelligence journal | New and Recent Articles</description>
        <language>en-us</language>
        <generator>Frontiers Feed Generator,version:1</generator>
        <pubDate>2026-05-08T19:09:30.948+00:00</pubDate>
        <ttl>60</ttl>
        <item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1786757</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1786757</link>
        <title><![CDATA[AQFormer: severity-aware transformer with aphasia-specific CAM for spoken keyword classification in aphasic speech]]></title>
        <pubdate>2026-05-08T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Gowri Prasood Usha</author><author>John Sahaya Rani Alex</author>
        <description><![CDATA[Language-driven speech output in individuals with aphasia shows considerable variability, including phonological errors and pauses during word searches. This makes it difficult to use traditional keyword classification systems and further reduces trust in deep neural models, complicating their application in clinical settings. This paper introduces AQFormer, a severity-aware transformer architecture designed to classify spoken keywords in aphasic speech, and A-CAM, a dual-stream attribute framework aimed at assisting individuals with aphasic impairments. AQFormer generates acoustic representations that are severity-adaptive by integrating patient-level Aphasia Quotient (AQ) scores through Feature-wise Linear Modulation (FiLM) and A-CAM. A-CAM consists of two main components: (i) a branch that influences WavLM convolutional features, a prediction-focused one, and (ii) a multimodal aphasia filter that captures pauses, phoneme variations, and interruptions at word boundaries, an impairment-focused branch. We introduce an adaptive perturbation and dual-filtering gradient scheme that enforces non-negative, mask-consistent attributions over time-frequency regions. Experiments utilizing a subset of AphasiaBank keywords (93 speakers, 960 recordings; training set expanded to 5,138) with rigorous speaker-disjoint evaluation indicate that AQFormer achieves approximately 96.61% accuracy (F1 = 96.8%) on previously unseen speakers. A-CAM consistently outperforms several Grad-CAM variants when deletion/insertion AUPC and ADCC metrics are employed. This results in stable, sparse explanations that reflect how aphasia is usually caused: Discriminates correct from incorrect productions with Cohen’s d = 2.05 (a massive effect size) and spatial localization of error regions with Intersection over Union (IoU) of 0.461 against phoneme boundaries. Montreal Forced Aligner meets the quantitative validation criteria for the aphasia filter. The impairment-focused A-CAM maps achieve an IoU of 0.712 against detected error regions, with a severity correlation that doubles from rho = −0.374 (base) to rho = −0.754 (filter-gated). By tightly coupling severity-aware modelling with aphasia-informed attributes, the proposed framework advances explainable learning systems for aphasia-affected speech without needing clinician-labelled training targets.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1819009</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1819009</link>
        <title><![CDATA[Learning instance-specific counterfactual models for continuous treatments using hypernetworks]]></title>
        <pubdate>2026-05-08T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Roger Pros</author><author>Jordi Vitrià</author>
        <description><![CDATA[In recent years, non-linear machine learning techniques have attracted increasing attention for estimating treatment effects from observational data. While most existing methods focus on binary treatment scenarios, the estimation of treatment effects for continuous interventions remains a critical challenge in many real-world applications. In this work, we introduce a novel neural network-based approach for estimating continuous treatment effects by leveraging hypernetworks to model counterfactual outcomes across treatment levels. This approach extends the principles of binary treatment effects computation to the continuous domain, addressing the key challenge of treatment relevance. By generating weights for a fixed network that predicts potential outcomes, our architecture ensures that the treatment variable retains its causal significance while maintaining the flexibility of deep learning models. Through extensive experiments on synthetic and semi-synthetic datasets, we demonstrate that our approach outperforms existing methods in terms of precision. The results highlight the advantages of explicitly modeling the relationship between treatment levels and outcomes, particularly in settings where traditional methods struggle with high-dimensional confounders or non-linear treatment-response dynamics.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1824060</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1824060</link>
        <title><![CDATA[Rethinking criminal profiling through cognitive artificial intelligence]]></title>
        <pubdate>2026-05-08T00:00:00Z</pubdate>
        <category>Mini Review</category>
        <author>Jorge Buele</author><author>Davis Miranda-Toapanta</author><author>David Rojas-Cañizares</author>
        <description><![CDATA[IntroductionTraditional criminal investigation often struggles to integrate dispersed and heterogeneous information, delaying the identification of serial or escalating patterns. Advances in artificial intelligence (AI) and cognitive computing offer data-driven approaches for cross-source correlation and temporal anomaly detection.MethodsA focused narrative review of peer-reviewed literature on AI applications in forensic analysis, pattern detection, and investigative support was conducted using major multidisciplinary databases. Selected studies were synthesized into two analytical dimensions: evidence correlation and anomaly detection, and further examined through a retrospective case-based illustration.ResultsAI-based approaches support the linkage of low-level traces with higher-level events, enabling structured reconstruction and large-scale pattern identification. Machine learning models integrate heterogeneous data into operational representations, achieving high predictive performance under controlled conditions, often reported above 90% accuracy. Statistical and learning-based methods also detect temporal compression, behavioral drift, and cross-source anomalies, revealing patterns that may remain fragmented under manual analysis. Retrospective examination of historically fragmented cases highlights longitudinal regularities, including shrinking inter-event intervals, increasing severity, and the accumulation of weak signals that become meaningful when analyzed jointly.DiscussionAI contributes primarily at a methodological level by enabling continuous integration and re-evaluation of investigative signals, supporting more data-informed and longitudinally grounded profiling practices. However, a gap persists between high performance in controlled settings and limited validation in real-world contexts. Future research should prioritize empirical benchmarking using operational datasets, the development of explainable and auditable systems, and governance frameworks ensuring transparent, accountable, and human-supervised deployment.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1817837</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1817837</link>
        <title><![CDATA[Spark: modular spiking neural networks]]></title>
        <pubdate>2026-05-08T00:00:00Z</pubdate>
        <category>Technology and Code</category>
        <author>Mario Franco</author><author>Carlos Gershenson</author>
        <description><![CDATA[Nowadays, neural networks act as a synonym for artificial intelligence. Present neural network models, although remarkably powerful, are inefficient both in terms of data and energy. Several alternative forms of neural networks have been proposed to address some of these problems. Specifically, spiking neural networks are suitable for efficient hardware implementations. However, effective learning algorithms for spiking networks remain elusive, although it is suspected that effective plasticity mechanisms could alleviate the problem of data efficiency. Here, we present a new framework for spiking neural networks—Spark (https://github.com/Nogarx/Spark)—built upon the idea of modular design, from simple components to entire models. The aim of this framework is to provide an efficient and streamlined pipeline for spiking neural networks. We showcase this framework by solving the sparse-reward cartpole problem with simple plasticity mechanisms. We hope that a framework compatible with traditional ML pipelines may accelerate research in the area, specifically for continuous and unbatched learning, akin to the one animals exhibit.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1795030</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1795030</link>
        <title><![CDATA[Hybrid Ant-Baby Optimizer and BiLSTM framework for high-performance IoT intrusion detection]]></title>
        <pubdate>2026-05-07T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Anbarasu Balakrishnan</author><author>Praveen Kumar Reddy Maddikunta</author>
        <description><![CDATA[IntroductionIntrusion detection in Internet of Things networks continuously encounters issues with high-dimensional traffic data, class imbalance, and patterns of attacks.MethodsThis paper proposes a hybrid intrusion detection framework which combines a novel bio-inspired Ant-Baby Optimizer for feature selection with a Bidirectional Long Short-Term Memory neural network for classification. The Ant-Baby Optimizer algorithm selects small, informative subsets of features using mutual information to replace features guaranteed to be low-relevance, thereby enhancing computational efficiency and interpretability. The selected features are re-shaped into pseudo-sequential order to allow, via the Bidirectional Long Short-Term Memory, to capture temporal dependencies in both directions.ResultsThe results on the CICIoT2023 dataset demonstrate superior detection accuracy (97.32%) balanced per-class precision, recall, and F1-scores (macro F1 = 0.9732), with an inference latency of 0.219 ms per sample and process time of over 4,500 samples per second.DiscussionExperiments and verification against state-of-the-art metaheuristic and deep learning approaches support the proposed framework as effective. The interdependence of computational complexity and dataset limitation cannot be overstated. In fact, these four qualities of performance, efficiency, interpretability, and adaptiveness, when considered as one, will pave the way for the proposed framework to perform at a high level in the real-world IoT intrusion detection systems. Moreover, future works will involve cross-domain validation, attention-based mechanisms, and online learning for zero-day attack handling.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1759110</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1759110</link>
        <title><![CDATA[ExplainTS: a benchmark dataset of pretrained models and post-hoc explanations for time-series classification]]></title>
        <pubdate>2026-05-05T00:00:00Z</pubdate>
        <category>Data Report</category>
        <author>Maciej Mozolewski</author><author>Szymon Bobek</author><author>Grzegorz J. Nalepa</author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1754973</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1754973</link>
        <title><![CDATA[AMULED: Addressing Moral Uncertainty using Large language models for Ethical Decision-making]]></title>
        <pubdate>2026-05-04T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Rohit K. Dubey</author><author>Damian Dailisan</author><author>Sachit Mahajan</author>
        <description><![CDATA[IntroductionWe address moral uncertainty in reinforcement learning (RL) by proposing a framework that integrates multiple ethical theories into decision-making. Existing approaches rely on single moral frameworks or handcrafted rewards, limiting scalability and failing to capture moral pluralism. We introduce AMULED, a task-agnostic ethical layer that refines a pre-trained RL agent using large language models (LLMs) to provide multi-perspective moral feedback.MethodsFollowing initial training, the RL model is fine-tuned using LLM-generated feedback in place of human feedback. Five moral clusters—consequentialist, deontological, virtue, care, and social justice—assign belief values to candidate actions. These beliefs are aggregated using Belief Jensen–Shannon Divergence and Dempster–Shafer Theory to produce probability scores that serve as shaping rewards, while a KL-regularization term constrains deviation from the base policy. The framework is evaluated across two environments (Finding Milk and Driving and Rescuing), multiple LLM backbones, and alternative belief aggregation methods, with 50-run replicates.ResultsAMULED improves ethical behavior without substantially degrading task performance. In Finding Milk, it increases desirable actions (63.1% more crying babies attended) and reduces undesirable actions (60.3% fewer sleeping babies disturbed), with only a 5.1% increase in path length. In Driving and Rescuing, it balances competing objectives more effectively than baselines, rescuing 38.4% more targets than human-feedback agents while maintaining lower collision rates and reduced policy degradation. Across experiments, BJSD-DST aggregation outperforms standard methods (e.g., voting, averaging) in handling conflicting moral signals and achieves the best overall performance on most metrics.DiscussionAMULED operationalizes moral pluralism through scalable, LLM-based feedback and provides a principled mechanism for resolving conflicting ethical signals. The framework demonstrates robustness across tasks and model variants, though performance depends on LLM reasoning quality and can degrade in spatially complex settings. These results suggest that LLM-driven belief aggregation offers a practical alternative to handcrafted rewards and human supervision for ethical decision-making in RL.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1767330</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1767330</link>
        <title><![CDATA[HyRA-CXR: a hybrid residual–attention deep network for chest X-ray classification]]></title>
        <pubdate>2026-05-04T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Ahmeed Suliman Farhan</author><author>Umar Manzoor</author><author>Ali Al-kubaisi</author><author>Ayman Jalil</author><author>Zeeshan Perveez</author><author>Ahmed B. Mohammd</author><author>Wadhah Zeyad Tareq</author>
        <description><![CDATA[Chest X-ray (CXR) interpretation is essential for diagnosing pulmonary diseases, yet manual reading remains slow and prone to human error, especially in high-volume or resource-limited settings. To address delayed diagnoses and improve clinical efficiency, this study introduces (HyRA-CXR), a hybrid residual–attention convolutional neural network for automated CXR classification. The proposed model integrates residual blocks to enhance gradient stability and dual attention mechanisms to focus on significant lung regions. Experiments are conducted on the publicly available Lung X-Ray Image dataset. Hyperparameters are optimized using KerasTuner as well as model evaluation is carried out using five-fold stratified cross-validation. HyRA-CXR achieved an average accuracy of 90.39%, outperforming DenseNet121 (89.38%) and Xception (89.12%) models. Also, the experimental results confirmed that both residual and attention modules contribute, as removing either reduced accuracy below 90%. Overall, the proposed model achieves competitive accuracy with maintaining a compact architecture (0.52M parameters), indicating its suitability for deployment in resource-constrained settings. Our code is publicly available at: https://github.com/Ahmeed-Suliman-Farhan/HyRA-CXR-A-Hybrid-Residual-Attention-Deep-Network-for-Chest-X-Ray-Classification.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1817216</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1817216</link>
        <title><![CDATA[Flood hazard assessment and zonal prioritization through an LR-bipolar triangular fuzzy hybrid decision-making approach]]></title>
        <pubdate>2026-05-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Ajeesh Puthusserry Paulose</author><author>Felix Augustin</author>
        <description><![CDATA[IntroductionFlood risk assessment has become increasingly important in regions vulnerable to climate-induced disasters. This study addresses the need for a robust decision-support framework by proposing a hybrid multi-criteria decision-making (MCDM) model to prioritize flood-prone zones in the Ernakulam district of Kerala, aligning with sustainable development goals focused on climate resilience.MethodsThe proposed approach employs LR-bipolar triangular fuzzy numbers (LRBTFNs) to effectively capture uncertainty in decision-making. It integrates three well-established MCDM techniques, Preference Selection Index (PSI), Simple Additive Weighting (SAW), and Combinative Distance-based Assessment (CODAS), to ensure a balanced and comprehensive ranking process. The methodology incorporates normalization, aggregation, and defuzzification steps to compute performance scores. Additionally, unsupervised learning techniques, namely K-means clustering and Principal Component Analysis (PCA), are utilized to validate vulnerability patterns and group regional profiles.Results and discussionThe results reveal that Kochi, Vypen, and Paravoor are the most vulnerable flood-prone zones, while Kothamangalam is identified as the least susceptible area. The integration of multiple MCDM methods enhances the robustness and reliability of the ranking outcomes, and clustering and PCA analyses further confirm consistent vulnerability trends across regions. The findings provide valuable insights for policymakers and local authorities to implement targeted risk mitigation and planning strategies. Moreover, the study supports Sustainable Development Goal 13 (Climate Action) by promoting resilience and preparedness against climate-induced flood hazards.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1794876</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1794876</link>
        <title><![CDATA[Feature stabilization in convolutional neural networks using Proportional Integral Controller for lung nodule classification]]></title>
        <pubdate>2026-04-30T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>V. Sangeetha</author><author>S. Kalaivani</author>
        <description><![CDATA[IntroductionReliable classification of lung nodules from computed tomography (CT) images remains a challenging problem due to variations in image intensity, noise, and unstable feature representations during deep network training. Although convolutional neural networks (CNNs) have achieved promising results in medical image analysis, their internal feature dynamics are often difficult to control, which can affect convergence stability and generalization, particularly when working with limited clinical data.MethodIn this work, we propose a control-inspired CNN framework that incorporates a Proportional Integral Controller (PIC) to regulate feature representations during the learning process. The PIC is integrated into the network in two distinct ways: as a preprocessing module before the CNN and as an intermediate layer embedded within the convolutional architecture. Both manually tuned and automatically learned PIC configurations are investigated to analyze the influence of fixed, knowledge-driven control parameters vs. adaptive, data-driven feedback mechanisms. The proportional component responds to instantaneous feature deviations, while the integral component compensates for accumulated errors, jointly contributing to more stable and consistent feature learning.ResultThe proposed approach is evaluated on the IQ-OTH/NCCD lung cancer dataset using standard classification metrics. The proposed method achieves state-of-the-art performance (Accuracy 0.96, F1-score 0.96) and eliminates false positives (Precision 0.93). Ablation and statistical analyses confirm the importance of PIC placement and parameter tuning, while cross-dataset validation demonstrates strong generalization. Overall, this study demonstrates that integrating principles from control theory into deep learning architectures provides an effective and interpretable strategy for enhancing medical image classification.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1857609</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1857609</link>
        <title><![CDATA[Correction: Retrieving interpretability to support vector machine regression models in dynamic system identification]]></title>
        <pubdate>2026-04-29T00:00:00Z</pubdate>
        <category>Correction</category>
        <author>Johan Pena-Campos</author><author>Diego Patino</author><author>Carlos Ocampo-Martinez</author><author>Julio C. Ramos-Fernández</author><author>Margot Salas-Brown</author><author>Alexander Caicedo</author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1807268</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1807268</link>
        <title><![CDATA[Efficient and accurate medical AI: MediLore and MediOut]]></title>
        <pubdate>2026-04-28T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>S. Mohamed Rayhan</author><author>M. Hariprasath</author><author>K. Hemalatha</author>
        <description><![CDATA[IntroductionThe integration of artificial intelligence (AI) in medical question-answering (QA) systems requires a careful balance between diagnostic accuracy and computational efficiency. Existing large language models (LLMs) achieve strong performance but are often limited by high memory usage, latency, and inconsistent behavior in handling rare or complex clinical queries. This study addresses these limitations by exploring efficient and robust modeling strategies for medical QA.MethodsTwo complementary approaches were developed: MediLore and MediOut. MediLore employs Weighted Low-Rank Adaptation (LoRA) adapter fusion to integrate domain-specific knowledge into a shared backbone model while reducing computational overhead. MediOut utilizes an output-level ensembling strategy that aggregates predictions from multiple fine-tuned models using semantic similarity-based scoring. Both models were trained and evaluated on clinically curated datasets, including MedQA, PubMedQA, and MedMCQA. Performance was assessed using BLEU, ROUGE, BERTScore, and BioBERT-based similarity metrics. Additionally, 4-bit quantization was applied to optimize deployment efficiency.ResultsMediOut achieved the highest performance across semantic evaluation metrics, with a BioBERT F1 score of 0.934 and strong improvements in semantic similarity and contextual alignment. MediLore retained up to 91% of the ensembling accuracy while reducing inference cost to approximately 0.3% of the baseline, significantly lowering latency from 141 seconds to 190 ms. BLEU score improvements were moderate (0.066-0.074), indicating that semantic alignment gains were more substantial than lexical overlap improvements.DiscussionThe results demonstrate that MediLore and MediOut provide complementary advantages in medical QA systems. MediLore enables efficient deployment in resource-constrained environments, while MediOut enhances robustness and semantic fidelity for complex clinical queries. The proposed framework highlights the trade-off between efficiency and accuracy, offering practical guidance for selecting appropriate deployment strategies in real-world healthcare applications. These findings contribute to the development of scalable, reliable, and clinically aligned AI systems for biomedical natural language processing.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1751148</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1751148</link>
        <title><![CDATA[CompoundDenseNet: a novel approach for accurate recognition of Bangla handwritten compound characters]]></title>
        <pubdate>2026-04-15T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Nazia Alfaz</author><author>Talha Bin Sarwar</author><author>Fahmid Al Farid</author><author>Md Saef Ullah Miah</author><author>Sadia Afrin</author><author>Shakila Rahman</author><author>Jia Uddin</author><author>Hezerul bin Abdul Karim</author>
        <description><![CDATA[Bangla, one of the most widely spoken languages in the world, presents major challenges in handwritten character recognition because of its complex compound characters with intricate shapes, diverse writing styles, and structural similarities. These features make Bangla a representative example of complex scripts that remain difficult for conventional Optical Character Recognition (OCR) systems. This study focuses on improving the recognition of Bangla handwritten compound characters using a modified DenseNet architecture named CompoundDenseNet. The architecture enhances feature extraction and reuse to better capture the visual variations and fine structural details that existing models often struggle to handle. Its performance was evaluated on three benchmark datasets, BanglaLekha Isolated, Ekush, and CMATERdb, achieving recognition accuracies of 98.5%, 98%, and 96.2% respectively, surpassing previously reported methods. Misclassification analysis using a confusion matrix revealed that the Adam optimizer produced the most stable and accurate results with faster convergence compared to other optimizers tested. While the results demonstrate significant progress, the study also highlights the need for larger and more diverse datasets. Overall, CompoundDenseNet contributes to advancing Bangla handwritten compound character recognition and has the potential to enhance real-world applications such as education, legal documentation, and digital accessibility in Bangla language technologies.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1816684</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1816684</link>
        <title><![CDATA[Editorial: Advancing AI-driven code generation and synthesis: challenges, metrics, and ethical implications]]></title>
        <pubdate>2026-04-15T00:00:00Z</pubdate>
        <category>Editorial</category>
        <author>Sumeet Kaur Sehra</author><author>Sukhjit Singh Sehra</author><author>David S. Allison</author><author>Jaiteg Singh</author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1800302</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1800302</link>
        <title><![CDATA[Delegated agency and moral responsibility in artificial intelligence]]></title>
        <pubdate>2026-04-15T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Petar Radanliev</author>
        <description><![CDATA[IntroductionArtificial intelligence ethics is often framed as a response to unprecedented technical autonomy, with risks attributed to recent advances in machine learning and scale. This framing overlooks a recurring ethical structure: the delegation of moral authority to artificial agents. Ethical failures associated with AI are best understood as governance failures rooted in human design choices and accountability arrangements, even where opacity and limited control complicate responsibility attribution.MethodsA qualitative, interdisciplinary approach integrates historical–thematic analysis, comparative interpretation of technological artifacts, and visual–conceptual synthesis. Mythological figures (Talos, the Golem, Pygmalion), early mechanical automata, and foundational computational systems are analyzed as conceptual models of delegated artificial agency rather than technological precursors.ResultsAcross historical contexts, artificial agents exhibit consistent structural features: bounded autonomy, delegated authority, explicit override mechanisms, and dependence on human oversight. These features directly correspond to contemporary AI ethics concerns, including alignment failures, responsibility gaps, human-in-the-loop control, and system interruptibility.DiscussionThe analysis establishes that ethical risk in AI arises from the displacement of human responsibility rather than from machine autonomy. By situating AI within a longer history of artificial agency, the study provides a normative framework that locates moral responsibility unambiguously in human actors and institutions, with direct implications for AI governance and accountability.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1842850</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1842850</link>
        <title><![CDATA[Correction: Evaluation of large language models in generating and optimizing educational materials for neonatal home oxygen therapy]]></title>
        <pubdate>2026-04-13T00:00:00Z</pubdate>
        <category>Correction</category>
        <author>Zhendong Liu</author><author>Xiaoping Yang</author><author>Yu Zhang</author><author>Yujing Xu</author><author>Yue Xiang</author><author>Hongyan Wang</author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1835185</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1835185</link>
        <title><![CDATA[Correction: The association between national culture and AI readiness: a cross-national study]]></title>
        <pubdate>2026-04-13T00:00:00Z</pubdate>
        <category>Correction</category>
        <author>Frontiers Production Office </author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1713747</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1713747</link>
        <title><![CDATA[Explainable neuro-symbolic artificial intelligence for automated interpretation of corneal topography and early keratoconus detection]]></title>
        <pubdate>2026-04-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Mini Han Wang</author><author>Shuai Qin</author>
        <description><![CDATA[BackgroundEarly detection of keratoconus is essential for preventing postoperative complications in refractive surgery and preserving long-term visual function. Although artificial intelligence has demonstrated strong potential in ophthalmic image analysis, many existing models operate as black-box systems and provide limited clinical interpretability. Transparent decision support is therefore critical for safe deployment of AI in clinical practice.MethodsWe propose an explainable neuro-symbolic framework for automated interpretation of corneal topography reports and refractive surgery eligibility assessment. The proposed system integrates multimodal feature extraction, a symbolic corneal knowledge graph, probabilistic reasoning, and large language model (LLM)–based report generation. Quantitative biometric parameters and corneal curvature maps extracted from IOLMaster 700 reports were processed through a hybrid convolutional neural network–Vision Transformer (CNN–ViT) module to capture spatial corneal morphology. These representations were aligned with a clinically curated knowledge graph encoding relationships between corneal parameters, disease states, and surgical decision criteria. Bayesian probabilistic inference was then applied to estimate disease likelihoods, while an ensemble LLM module generated structured bilingual clinical reports explaining the reasoning process.ResultsIn a prospective pilot cohort of 20 eyes, the proposed framework demonstrated strong diagnostic performance for early keratoconus detection, achieving an area under the receiver operating characteristic curve (AUC) of approximately 0.95. Sensitivity and specificity remained high across decision thresholds, and the system achieved a balanced F1 score for refractive surgery eligibility classification. Expert evaluation indicated high interpretability and clinical usefulness of the generated reports. The end-to-end pipeline required approximately 95 ± 12 s per case, supporting near–real-time clinical decision support.ConclusionThe proposed neuro-symbolic framework combines deep representation learning, structured medical knowledge, and explainable language-based reporting to provide transparent AI-assisted corneal diagnostics. Although the current results are based on a pilot cohort, the framework demonstrates the potential of integrating neural networks, knowledge graphs, and large language models for interpretable ophthalmic AI systems. Future studies using larger multicenter datasets are required to further validate clinical performance and generalizability.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1777258</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1777258</link>
        <title><![CDATA[Functional stability assessment and adaptation for critical infrastructure facilities]]></title>
        <pubdate>2026-04-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Victor Perederyi</author><author>Eugene Borchik</author><author>Viacheslav Zosimov</author><author>Oleksandra Bulgakova</author>
        <description><![CDATA[IntroductionEnsuring functional stability of critical infrastructure facilities (CIFs) under conditions of uncertainty and dynamic threats remains a critical challenge. Existing approaches insufficiently integrate technical, cybersecurity, and human-related factors.MethodsThis study proposes an information-cognitive approach based on a hybrid model combining Bayesian Trust Networks and fuzzy logic. The model incorporates expert knowledge and evaluates the mutual influence of information security, cybersecurity, human factors, and vulnerability indicators. The Mamdani algorithm is used for probabilistic estimation under uncertainty.ResultsNumerical experiments conducted in the GeNIe environment demonstrate that the proposed model effectively supports decision-making. Scenario analysis shows that adjusting key cybersecurity and vulnerability factors increases the probability of achieving sufficient functional stability above the critical threshold.DiscussionThe proposed hybrid framework improves interpretability and adaptability of functional stability assessment. It enables flexible reasoning under uncertainty and supports real-time decision-making for critical infrastructure management. The approach can be applied across different categories of CIFs and extended with additional data-driven components.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1792860</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1792860</link>
        <title><![CDATA[Heart disease prediction using rough neutrosophic sets and dual-attention neural networks: RNS-OptiDANet]]></title>
        <pubdate>2026-04-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>T. Ashika</author><author>G. Hannah Grace</author>
        <description><![CDATA[IntroductionHeart disease is a major global health problem that highlights the need for effective and accurate prediction methods.MethodsThis paper presents RNS-OptiDANet, a hybrid framework that combines rough set theory (RST), rough neutrosophic sets (RNS) and an optimized dual-attention neural network (OptiDANet) in order to predict heart disease. For feature selection, the QuickReduct method with the discernibility matrix (RST QRDM) was used. The features selected in RST were represented as RNS representations to deal with uncertainty in the classification process. The OptiDANet model implements Dual Attention Mechanisms such as Channel Attention (CAM) and Soft Attention Mechanism (SAM) to highlight the relevant patterns while reducing noise. The performance of the developed framework has been improved through Hyperparameter tuning using Optuna and overfitting has been avoided. Finally, classification is conducted using a Random Forest (RF) model.ResultsExperimental results demonstrate strong performance in terms of accuracy, precision, recall and F1-score across datasets.DiscussionAn eXplainable Artificial Intelligence (XAI) module is integrated to provide feature level interpretability and clinical transparency while ablation study validates the contribution of each framework component confirming the robustness and effectiveness of the proposed hybrid RNS-OptiDANet model.]]></description>
      </item>
      </channel>
    </rss>