<?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 | New and Recent Articles</title>
        <link>https://www.frontiersin.org/journals/artificial-intelligence</link>
        <description>RSS Feed for Frontiers in Artificial Intelligence | New and Recent Articles</description>
        <language>en-us</language>
        <generator>Frontiers Feed Generator,version:1</generator>
        <pubDate>2026-09-06T14:26:48.908+00:00</pubDate>
        <ttl>60</ttl>
        <item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1852439</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1852439</link>
        <title><![CDATA[Predicting first falls among older adults with chronic conditions and polypharmacy using routinely collected health records]]></title>
        <pubdate>2026-09-04T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Igor Larrañaga</author><author>Miguel Rujas</author><author>Itxaso Alayo</author><author>Irati Erreguerena</author><author>Marco Capo</author><author>Laura Lopez-Perez</author><author>Marta M. Mediavilla</author><author>María Dolores Martínez</author><author>Héctor Chuliá-Gil</author><author>Eleni I. Georga</author><author>Muhammad Salman Haleem</author><author>Leandro Pecchia</author><author>Dimitrios I. Fotiadis</author><author>Giuseppe Fico</author><author>Ane Fullaondo</author>
        <description><![CDATA[BackgroundFalls represent a major clinical and financial challenge for healthcare systems. Accurately predicting first falls remains challenging, especially when using routinely collected data.ObjectiveDevelop and evaluate a predictive model to identify elderly people at risk of first fall—defined as a fall after a 90-day fall-free period—in the Basque Country using routinely collected health records.MethodA retrospective study included patients aged ≥65 with at least two chronic conditions among heart failure, chronic obstructive pulmonary disease (COPD), and diabetes. Data on demographics, diagnoses, prescriptions and healthcare utilisation were obtained from Osakidetza-Basque Health Service databases. Patients were labelled as “fallers” if they fell during 2022–2023 after a 90-day fall-free period rather than a true first-ever fall. Predictive models—logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB)—were trained using recursive feature elimination with cross-validation (RFECV). Shapley additive explanations (SHAP) enhanced model interpretability and explainability.Results35,197 patients were included, with 10.6% experiencing a fall. All models achieved similar results, with an AUCROC score of 0.71, while precision remained low. Emergency room visits in the prior 3 months, presence of caregiver, and age consistently ranked the top predictors. The number of prescriptions and antidepressant use also emerged as relevant.DiscussionModels showed moderate predictive performance. Relying solely on routinely collected health records limited clinical applicability due to low precision. Future work should integrate diverse data sources—health records, real-time gait and balance metrics, environmental factors—to improve fall prediction and support clinical decision-making.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1840721</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1840721</link>
        <title><![CDATA[A robust privacy-preserving federated framework for kidney CT image classification using transfer learning models]]></title>
        <pubdate>2026-09-04T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Sai Sri Hemantha Konala</author><author>Srinivas Koppu</author>
        <description><![CDATA[IntroductionKidney abnormalities, including cysts, tumors, and stones, are the most common renal disorders that can lead to severe complications such as chronic kidney disease or renal failure. Deep learning-based medical image analysis offers an effective approach for the accurate classification of kidney abnormalities, aiding the early diagnosis of renal disorders. However, its centralized training leads to inadequate privacy protection.MethodsConsidering the importance of ensuring individuals' data privacy, this study proposes a novel federated transfer learning framework for accurate classification of renal abnormalities using 12,446 kidney CT scan images and simultaneously preserves data privacy. CT scan images were preprocessed by resizing and normalization, followed by data augmentation techniques, including random rotations (±30°), horizontal flips, and color jitter, to address class imbalance and improve model generalization. Five pre-trained deep learning models such as MobileNetV2, EfficientNetV2-S, ResNet50, DenseNet121, and InceptionResNetV2 were trained across seven federated clients. Federated weighted averaging was employed for aggregation, and AES-256 encryption in CBC mode was applied to all model parameter transmissions between clients and the server.ResultsMobileNetV2 achieved the best performance, attaining 99.48% accuracy, 99.29% precision, 99.32% recall, 99.3% F1-score, 0.9999 AUC-ROC, and log loss of 0.0247. Cross-client validation produced an average accuracy of 98.85% with a generalization gap of only −0.0063, indicating strong generalization across client datasets.DiscussionThe proposed framework provides an effective balance between privacy preservation and communication efficiency, highlighting its potential for deployment in distributed clinical environments for kidney disease diagnosis.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1883853</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1883853</link>
        <title><![CDATA[An agentic AI framework connecting language models to electronic health records and a biomedical knowledge graph for real-world evidence]]></title>
        <pubdate>2026-09-04T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Gianmarco Bellucci</author><author>Wanjun Gu</author><author>Peter W. Rose</author><author>Sergio E. Baranzini</author>
        <description><![CDATA[BackgroundAccessing large-scale clinical and biomedical databases remains a significant barrier for clinicians and researchers, requiring substantial computational expertise. Agentic artificial intelligence frameworks, in which large language models (LLMs) orchestrate multi-step reasoning and query execution under interactive human supervision, offer the potential to democratize data access and accelerate evidence generation.MethodsWe applied the Model Context Protocol (MCP) to integrate, within a single agentic workflow, an Observational Medical Outcomes Partnership (OMOP)-standardized electronic health record (EHR) database from an academic health system (>7 million subjects), with the Scalable Precision Medicine Open Knowledge Engine (SPOKE), a curated knowledge graph integrating relationships biomedical concepts from expert-maintained resources. Using this implementation (MedCP), we evaluated the approach across 100 benchmarking clinical research tasks, 617 biomedical factual accuracy questions (BiomixQA), an integrative case study linking clinical co-occurrence with molecular similarity, and a standardized protocol for generating and replicating real-world studies.ResultsAcross the 100-task benchmark, knowledge-graph access improved mean scores overall for GPT-5.5 and Claude Opus 4.8, though the pattern differed between models; on BiomixQA, SPOKE grounding raised multiple-choice accuracy for both models, without changing true/false accuracy. In the case study, disease co-occurrence patterns extracted from the EHR correlated with molecular network similarity, surfacing mechanistic hypotheses from real-world data. Applied to the replication of published observational studies, the research protocol compressed timelines from months to hours, lowering the technical barrier to query generation and execution while study design and interpretation remained under expert supervision.ConclusionsAn agentic AI infrastructure that combines institutional EHR data with curated biomedical knowledge via MCP can serve as a transparent, domain-grounded, supervised research assistant for real-world evidence, supporting both hypothesis generation and testing.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1965611</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1965611</link>
        <title><![CDATA[Correction: Editorial: Digital medicine and artificial intelligence]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Correction</category>
        <author>Mini Han Wang</author><author>Norberto Peporine Lopes</author><author>Nuno S. Osório</author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1848787</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1848787</link>
        <title><![CDATA[Deep convolutional neural network-based automated identification and classification of mungbean foliar diseases]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Shail Bala</author><author>S. I. Harlapur</author><author>Aditya Kamalakar Kanade</author><author>M. P. Potdar</author><author>Gurupada B. Balol</author><author>V. B. Kuligod</author><author>Lingareddy Usha Rani</author>
        <description><![CDATA[Early and accurate detection of plant diseases is critical in precision agriculture to improve crop management and yield. Mungbean (Vigna radiata L.) is highly susceptible to several foliar diseases, including yellow mosaic, powdery mildew, leaf crinkle, and cercospora leaf spot, which cause substantial productivity losses. Despite expanding applications of deep learning in plant disease diagnosis, systematic multi-architecture evaluation for mungbean disease classification under natural field conditions remains limited. This study addresses this gap by evaluating five state-of-the-art deep convolutional neural network (DCNN) architectures on a large-scale, field-acquired mungbean dataset that captures real-world variability across environmental conditions and disease severity levels, distinguishing it from controlled laboratory studies. A total of 5,617 original images across five classes were used. Data augmentation was applied exclusively to the training subset after stratified splitting to prevent data leakage. The dataset was partitioned into training (70%), validation (15%), and testing (15%) subsets. VGG16, VGG19, ResNet50V2, DenseNet121, and InceptionV3 were evaluated using identical transfer learning and fine-tuning protocols. Model performance was assessed using AUC-ROC, Cohen's kappa coefficient, McNemar's test for pairwise statistical comparisons, five-fold cross-validation, and Grad-CAM-based interpretability. On the independent test set, InceptionV3 achieved the highest accuracy (98.47%) and macro-F1 (98.49%), followed by VGG16 (98.36%) and VGG19 (97.89%). AUC-ROC values exceeded 0.997 for all models, confirming excellent class discrimination. Grad-CAM visualizations further confirmed that model predictions were based on biologically relevant disease symptoms. The findings demonstrate the effectiveness of deep learning for robust disease recognition under realistic field conditions and highlight the potential of AI-based diagnostic tools for crop health monitoring, precision agriculture, and decision-support systems in mungbean production.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1904923</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1904923</link>
        <title><![CDATA[The impact of artificial intelligence with multi-head self-attention based deep learning model for students' academic performance monitoring]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Mahmoud Ragab</author>
        <description><![CDATA[IntroductionEducational data mining has been applied to examine students' data collected from different educational organizations to forecast academic students' performance, which could assist them in accomplishing improved results in their upcoming courses.MethodsThis study presents the impact of artificial intelligence with multi-head self-attention on students' academic development using metaheuristic optimization algorithms (IAIMSA-SADMOA) model. The study aims to advance an effective method for students' academic performance using artificial intelligence tools to improve learning outcomes, engagement, and personalized education. The min-max normalization is employed in the data normalization phase for transforming input data into a beneficial format. The fruit fly optimization algorithm is deployed for the feature selection process to select the most related features from a dataset. Moreover, the proposed model designs bidirectional long short-term memory and multi-head self-attention mechanisms for the student's academic performance classification process. The parameter tuning process is performed through a sparrow search algorithm to enhance the classification performance.Results and DiscussionThe experimental evaluation of the IAIMSA-SADMOA technique occurs using a benchmark dataset. The empirical results indicated the enhanced performance of the proposed method in comparison with recent approaches.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1920911</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1920911</link>
        <title><![CDATA[AI-powered retinal imaging for early ASCVD risk stratification in a hybrid care model: preliminary findings from SAFER]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Hala Zakaria</author><author>Juman Ali</author><author>Mohammed Gouda Ibrahim</author><author>Idalys Roman</author><author>Ali Hashemi</author><author>Ihsan Almarzooqi</author>
        <description><![CDATA[BackgroundAI-enabled retinal imaging provides a non-invasive method for assessing cardiovascular risk by identifying microvascular changes linked to atherosclerotic cardiovascular disease. This study assessed the feasibility and clinical associations of the AI-derived retinal risk score (Reti-CVD) in a hybrid care model and compared its categorical alignment with the American Heart Association PREVENT equation as a comparator risk model.MethodsIn this retrospective cross-sectional study, 1,461 adults aged ≥40 at Metabolic underwent non-mydriatic retinal imaging. Fundus photographs were analyzed using the Dr. Noon AI platform to classify Reti-CVD risk. Associations with cardiometabolic characteristics were examined, and categorical agreement with PREVENT was assessed using Cohen’s kappa. ROC analyses evaluated the alignment of the continuous Reti-CVD score with PREVENT thresholds.ResultsHigher Reti-CVD risk was significantly linked to older age, male sex, diabetes, hypertension, elevated HbA1c, higher blood pressure, lower HDL cholesterol, higher triglycerides, and older AI-predicted vascular age (all p < 0.001). After adjustment, age, female sex, and current smoking were independent predictors. The Reti-CVD score showed moderate discrimination for PREVENT-defined risk thresholds (AUC 0.689–0.757) and slight to fair agreement with PREVENT (κ = 0.144; weighted κ = 0.241). Among patients classified as low risk by PREVENT, 49.1% were categorized as moderate or high by Reti-CVD.ConclusionAI-enabled retinal imaging was feasible in a hybrid setting and was associated with clinically plausible cardiometabolic risk gradients. Reti-CVD showed limited categorical agreement with PREVENT. However, the findings are hypothesis-generating and require prospective validation against ASCVD events before claims, improved prediction, or beneficial reclassification.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1832131</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1832131</link>
        <title><![CDATA[Advanced security framework for 5G-SDN attack detection and mitigation using Affinity Cohesive Clustering and Improved Chaotic Walrus-Optimized Convolutional LSTM with Deep Dual Info-Varcoder]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Shameli R.</author><author>Sujatha Rajkumar</author>
        <description><![CDATA[IntroductionThe integration of 5G and Software-Defined Networking (SDN) has introduced new security challenges due to limited processing capabilities, inefficient resource allocation, frequent user mobility, and the rapid growth of Internet of Things (IoT) devices. These dynamic network conditions increase the likelihood of malicious activities, highlighting the need for robust and intelligent intrusion detection techniques.MethodsThis research introduces a hybrid intrusion detection framework, termed APCHC-DDST-IV-CLSTM-ICWO, which integrates Affinity Propagated Cohesive Hierarchical Clustering, Deep Dual Stream SpaTemp Info-Varcoder, Dense Convolutional Long Short-Term Memory (CLSTM), and Improved Chaotic Walrus Optimization. The clustering module identifies attack patterns, while the dual-stream Info-Varcoder extracts representative spatial and temporal features. The Dense CLSTM captures sequential dependencies in network traffic to improve attack classification, and the Improved Chaotic Walrus Optimisation algorithm optimises the model parameters to enhance predictive performance. The proposed approach was validated using a 5G-SDN dataset generated with the MATLAB SDN Toolbox, with network traffic and attack scenarios simulated through the NS-3 network simulator.ResultsThe model showed a superior classification performance with a high average accuracy of 0.985±0.002, precision of 0.984±0.003, and F1 score of 0.985±0.003. The model’s low log-loss of 0.411 allows for the accurate separation of attack categories, including unknown attacks. This method improves the security and performance of 5G-SDN by increasing the throughput to 3218 samples/second and the detection time to 1.74 seconds.DiscussionThe results of the experiment demonstrate that the proposed APCHC-DDST-IV-CLSTM-ICWO framework offers a feasible method to secure 5G-SDN networks. The system provides high detection accuracy, effective processing, and reliable identification of emerging cyber threats by combining hierarchical clustering, deep feature learning, sequence modeling, and optimization algorithms. These findings indicate that it is appropriate for improving the operational effectiveness and security of next-generation 5G-SDN configurations.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1903342</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1903342</link>
        <title><![CDATA[Multi-resolution perceiver-based residual attention for explainable fault diagnosis in power transmission systems]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Shantanu Dinesh Wani</author><author>S. Abinaya</author><author>S. Abirami</author><author>R. Priyadarshini</author>
        <description><![CDATA[Fault diagnosis in power transmission systems requires the simultaneous identification of fault type and location from high-dimensional, multichannel, and temporally evolving measurements. Conventional deep learning approaches may struggle to jointly capture localized transient patterns, global temporal dependencies, and spatially distributed fault signatures while providing interpretable diagnostic information. To address these challenges, a Multi-Resolution Perceiver-Based Residual Attention (MRP-RA) framework is proposed, that combines temporal convolutional feature extraction with a learnable frequency-aware projection branch, Perceiver-based latent modeling, residual spatio-temporal attention, and coupled multi-task learning for fault detection, classification, and localization. The framework processes 256-step windows of 42-channel measurements from a simulated IEEE 5-bus transmission system, where the temporal and frequency-aware representations are fused before being compressed into a latent representation using 32 learnable tokens. Residual temporal and channel attention mechanisms are employed to identify salient temporal regions and feature channels associated with fault signatures, while the multi-task objective jointly optimizes anomaly detection, fault-type classification, and fault localization. Experimental results show that MRP-RA achieves 96.47% fault-type classification accuracy and 97.64% fault-location accuracy across multiple random seeds, while achieving 3.5 × higher inference throughput than the Transformer baseline under the evaluated configuration. These results demonstrate that the proposed combination of multi-resolution feature extraction, latent global modeling, and residual attention can provide accurate joint fault diagnosis with reduced attention-related computational requirements and interpretable attention-based diagnostic insights. The findings support MRP-RA as a promising approach for explainable fault diagnosis in simulated power transmission systems, while further validation across different grid topologies, operating conditions, and real-world measurements is required to establish broader generalizability.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1815425</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1815425</link>
        <title><![CDATA[Words matter: measuring and titrating the communicative character of AI]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>William C. Kouns</author>
        <description><![CDATA[As large language models become primary communication partners, the stylistic and communicative character of their output—specifically the degree to which it relies on intellectual, affective, or action-oriented language—shapes how users interpret, and act on what they read. Yet this property is rarely measured or controlled directly. We introduce the Intellect-Emotion-Action Profile (IEAP), a purpose-built lexical framework featuring an inductively constructed dictionary from AI-generated text. IEAP decomposes any response into the proportional usage of intellectual, affective, and action words. Using a single automated harvester, we elicited and scored responses from four contemporary architectures (Claude, ChatGPT, Grok, and Gemini) across five question domains under four instructional-mode directives that span a cold-to-fire register (n = 30 per architecture per domain). Three primary findings emerge. First, when no directive is applied, native communicative profiles are dominated by the underlying question, with architecture playing a secondary, architecture-dependent role; thus, the prompt establishes the baseline for the response. Second, an explicit mode directive displaces this profile substantially from its baseline. Specifically, the directives produced large, monotonic shifts in affective word usage, exceeding 50 percentage points in the most responsive domain with no reversals. Third, response depth alters length and conceptual breadth while leaving lexical composition essentially unchanged, establishing register and depth as orthogonal controls. These three dimensions correlated with human-rated NRC lexicons in the expected directions, supporting their validity. Within-domain robustness checks reproduce both the baseline profiles and the titration trajectories across multiple question framings. Because a measurable communicative property can be monitored and audited, these results position IEAP as a practical foundation for studying, comparing, and ultimately governing how AI systems communicate.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1795751</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1795751</link>
        <title><![CDATA[A severity-aware multi-task CNN-transformer framework for explainable diabetic foot ulcer triage and mobile deployment]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author> Abdullah</author><author>Muhammad Ateeb Ather</author><author>Zulaikha Fatima</author><author>José Luis Oropeza Rodríguez</author>
        <description><![CDATA[IntroductionDiabetic foot ulcer (DFU) image classification can support timely clinical triage; however, many existing systems provide limited severity granularity, rely on single-source image datasets, and predominantly use pixel-level explanation methods. This study developed a seven-class, severity-aware, multi-task artificial intelligence framework for interpretable DFU image analysis and smartphone-class deployment.MethodsThe framework combines domain-adaptive SimCLR pretraining, U-Net-based image refinement, EfficientNet-B0 feature extraction, a lightweight windowed transformer fusion block, and parallel classification and ordinal severity-regression heads. Four public data sources—DFUC 2020/2021, AZH, Medetec, and an IWGDF-aligned ordinal repository—were harmonized and deduplicated, yielding an analytical corpus of 24,925 original images. Evaluation included a 3,739-image held-out test set, five-fold stratified cross-validation, leave-one-source-out testing, an independent prospective clinician-audited cohort of 366 images, calibration and uncertainty analyses, concept-based interpretability assessment, and mobile-device profiling.ResultsOn the held-out test set, the framework achieved 95.5% accuracy (95% CI, 94.7–96.1) and a macro F1 score of 0.953, while mean five-fold cross-validation accuracy was 96.5% ± 0.3 percentage points. External ordinal severity estimation yielded a Spearman correlation of ρ = 0.91. In the clinician-audited cohort, weighted Cohen's κ for model-clinician severity agreement was 0.952. Model-assisted review was associated with a change in recorded management intent in 27.6% of cases, and 92.6% of generated explanations were rated as clinically useful. Full on-device inference, including gradient-weighted class activation mapping and concept-based explanation generation, required 322 ms per image on a Snapdragon 8 Gen 1 device.DiscussionThe proposed framework demonstrates how multi-source representation learning, multi-task severity modeling, uncertainty assessment, concept-based interpretability, and efficient edge deployment can be integrated within a unified AI pipeline for DFU image analysis. The findings indicate high classification performance, strong ordinal severity agreement, clinically interpretable outputs, and technically feasible smartphone-class inference. Further multicenter prospective validation across diverse populations and acquisition conditions is required before translation into routine clinical decision-support settings.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1886470</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1886470</link>
        <title><![CDATA[A comparative evaluation of deep learning models for the classification of encoded splice-junction sequences]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Yogesh Kumar</author><author>Inderpreet Kaur</author><author>Nandini Modi</author><author>Priya Bhardwaj</author><author>Jaeyoung Choi</author><author>Muhammad Fazal Ijaz</author>
        <description><![CDATA[IntroductionComputational classifiers can be benchmarked against public encoded splice-junction datasets, but the performance estimate will depend on the input representation, preprocessing pipeline, model set-up and evaluation design.MethodsThis study investigates the performance of 15 different Convolutional, Hybrid and Deep Learning architectures on a public encoded splice-junction dataset for 3-class classification. The results of the primary models were archived from one train/validation/held-out-test workflow and are presented as illustrative effects and not as statistically significant evidence of superiority.ResultsThe highest held-out test accuracy (96.87%) was obtained with the Custom CNN from among the configurations that were saved. The MLP Mixer was the most highly accurate model during training (99.90%), but the model did not perform as well on the validation set (93.33%) or held-out test set (93.73%), suggesting some in-sample fitting. Model-capacity information, parameter-to-sample ratios, learning curves, class-wise metrics, and a targeted class-weighting comparison are reported. Class-weighting was performed during the training phase only, but class-weighting did not yield highest performance for the tested Custom CNN configuration. of the chosen Custom CNN configuration for the observed dataset.DiscussionThe conclusions are limited to the classification results that could be computed for the encoded benchmark dataset and the reported evaluation procedure. The results do not demonstrate a statistically significant superiority of the models in the sense of external generalization, clinical diagnostic validity, patient-based mutation detection, clinical utility or readiness for deployment. Leakage-free repeated evaluation, external datasets, stored sample predictions, statistically comparison of models’ outputs, and clinically validated patient-level data would be needed for more general conclusions.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1867313</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1867313</link>
        <title><![CDATA[Robust vision transformer adaptation for UAV imagery]]></title>
        <pubdate>2026-09-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Jagan Murugesan</author><author>P. Helen Vijitha</author><author>L. Jai Vinita</author><author>A. K. Parvathy</author><author>P. N. Jeipratha</author>
        <description><![CDATA[Vision Transformers (ViTs) perform well on clean aerial imagery but degrade sharply when deployed in post-earthquake UAV operations, where motion blur, dust haze, illumination variation, and sensor noise combine to produce what we term post-earthquake visual drift, a structured distributional shift that can render an otherwise capable model dangerously unreliable in the field. Retraining or conventional domain adaptation is not a realistic option under the latency, compute, and annotation constraints of active disaster response. This work presents a comprehensive empirical evaluation of lightweight label-free test-time adaptation (TTA) methods for improving the robustness of ViTs under such deployment conditions, systematically comparing consistency-based self-supervision (MEMO) and entropy minimization (TENT) across datasets with varying classification complexity. Consistent with existing lightweight TTA approaches, only the LayerNorm affine parameters are adapted during inference, while the backbone and attention weights remain frozen. On the UAV-TEBDE post-earthquake dataset, accuracy rises from 73.47% under severe drift (severity 0.3) to 89.18% using MEMO, corresponding to a recovery of 15.71 percentage points over the drifted baseline, without using any ground-truth labels or performing any retraining. On UAV-TEBDE, MEMO also achieved higher recovery than the SAR baseline evaluated in this study. Additional evaluation on AID (30-class aerial scene classification) and UC Merced (21-class land-use classification) shows consistent drift-induced degradation and adaptation-driven recovery across datasets of varying complexity. The cross-dataset evaluation further reveals that the relative effectiveness of lightweight TTA methods depends on the classification-space dimensionality. TENT-based entropy minimization outperforms consistency-based MEMO when the class count is large, while MEMO is the stronger choice in low-class settings. This interaction between method design and classification-space dimensionality has direct consequences for choosing TTA strategies in operational disaster assessment pipelines. Feature-space PCA analysis confirms that LayerNorm-only updates geometrically recalibrate internal representations, rather than simply correcting the model's output logits.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1909141</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1909141</link>
        <title><![CDATA[NEUROCOG-DysNet: a multi-scale cognitive feature engineering framework with stacking ensemble for improved dyslexia screening]]></title>
        <pubdate>2026-09-02T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Sudha M</author><author>Murali S</author>
        <description><![CDATA[IntroductionDyslexia affects around 10% of the global population, yet early screening depends heavily on resource-intensive clinical assessments that are difficult to scale. Gamified cognitive platforms offer a practical alternative, but most existing machine learning approaches rely on raw task metrics without exploiting the hierarchical cognitive structure of multi-task assessment batteries.MethodsNEUROCOG-DysNet is a multi-scale cognitive feature engineering and stacking ensemble framework for dyslexia screening using gamified assessment data from the Dytective platform (N = 3,644; 392 dyslexic, 3,252 non-dyslexic). The framework extracts 258 hierarchical features across five cognitive levels-task, phase, temporal, global, and cross-domain-from raw game interactions, supplemented by four demographic variables. LightGBM-based feature selection retains the 50 most discriminative features. Seven base classifiers, configured to emphasize high sensitivity, balanced performance, or high AUC, are combined using five ensemble strategies and evaluated under a leakage-safe nested cross-validation protocol.ResultsThe stacking meta-learner achieves the highest composite score and attains an AUC-ROC of 0.8924 (95% CI: 0.8543–0.9253), sensitivity of 0.8077, specificity of 0.8187, and balanced accuracy of 0.8132. DeLong's test (p = 0.6946) and McNemar's test (p = 0.1443) indicate that AUC gains over the strongest individual baseline (XGB_HighAUC) are not statistically significant. Nested cross-validation confirms a small optimism gap (+0.011 AUC). Clinical utility analysis yields a 5.25 × lift in the highest-risk decile and a negative predictive value of 0.9726. Ablation shows that demographic variables contribute more to discrimination than any engineered cognitive level; a demographics-free variant attains AUC 0.7948.DiscussionNEUROCOG-DysNet provides a tunable sensitivity-specificity balance and well-calibrated risk scores suitable for a first-stage screening aid. The ensemble's advantage over fixed-weight alternatives is marginal on this dataset, and the cross-domain feature level is effectively inert. External validation across independent language cohorts and prospective multi-center evaluation are required before clinical deployment.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1902018</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1902018</link>
        <title><![CDATA[Evaluating hybrid quantum–classical learning for financial forecasting: a comparative study with classical machine learning models on BSE indices]]></title>
        <pubdate>2026-09-02T00:00:00Z</pubdate>
        <category>Review</category>
        <author>Rashmi Nair</author><author>Pankaj Chandre</author><author>Shahin Makubhai</author><author>Bhagyashree Shendkar</author><author>Disha Deotale</author><author>Vijaya Patil</author>
        <description><![CDATA[The forecasting of financial time series has gained more significance in decision making within a dynamic economic setting. Over the last few years, there has been a push in exploring both classical machine learning methods and novel quantum machine learning models with a view to enhance predictive accuracy. This paper suggests a quantum-classical regression framework developed that combines the feature selection in the form of a Random Forest in a hybrid approach, dimensionality reduction in the mode of Principal Component Analysis (PCA), and predicting in the form of a Variational Quantum Circuit (VQC). It is tested in various financial indicators, as BSE ESG, Carbon Exchange, Energy, Green Exchange, Oil and Gas and Power, with the help of RMSE and R2 as the measure of performance. As experimental findings indicate, the classical algorithms like Random Forest and k-Nearest Neighbors always have the best accuracy and stability compared to quantum and hybrid ones. Quantum models are unstable, in contrast to Quantum kNN which is fairly stable. The hybrid model performs moderately with consistent error values but fairly lower explanatory power with significant results mostly on the BSE ESG data. In general, the paper has pointed out that classical approaches are still best in structured financial prediction tasks and hybrid quantum-classical models are potentially effective, but not yet competent. The results give valuable information on the existing limitation and future of the quantum machine learning to financial forecasting applications.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1961460</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1961460</link>
        <title><![CDATA[Correction: The collaboration code: how humans and AI work together across millions of conversations on job tasks]]></title>
        <pubdate>2026-09-02T00:00:00Z</pubdate>
        <category>Correction</category>
        <author>Frontiers Production Office </author>
        <description></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1898557</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1898557</link>
        <title><![CDATA[Digital nudging in software development: a review and research agenda]]></title>
        <pubdate>2026-09-01T00:00:00Z</pubdate>
        <category>Review</category>
        <author>Ingo Pribik</author><author>Alexander Felfernig</author>
        <description><![CDATA[Digital nudging, a behavioral approach that subtly guides decision-making, is gaining increasing attention in software development as a means to support developers in complex and cognitively demanding tasks. This article analyzes the evolution of digital nudging and persuasive technology in software engineering from 2010 to April 2026. Based on a structured literature review, a temporal heatmap and qualitative trend analysis are used to examine how established nudging strategies, such as feedback, framing, and personalization, have been applied to support software development activities. The results demonstrate a shift from rule-based and socially oriented interventions toward more adaptive, context-aware, and AI/LLM-mediated approaches. In particular, recent work shows an increasing integration of personalized and real-time nudging mechanisms directly into development environments. Building on these findings, the article introduces a harmonized perspective on digital nudging, in which multiple strategies are systematically integrated to provide consistent and context-aware decision support. An illustrative AI-driven plug-in prototype demonstrates how personalized nudging mechanisms can be integrated into a development environment to support software quality and adherence to design principles such as SOLID. Furthermore, key challenges are identified, including the selection and combination of nudging strategies, their evaluation, and their integration into real-world development workflows. Overall, the article contributes a structured synthesis of existing research and outlines directions for future work on harmonized, AI-driven nudging in software engineering.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1891738</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1891738</link>
        <title><![CDATA[Synthetic resonance: a framework for growth-oriented human–AI relationships]]></title>
        <pubdate>2026-09-01T00:00:00Z</pubdate>
        <category>Conceptual Analysis</category>
        <author>Richard A. Fabes</author>
        <description><![CDATA[As human relationships with artificial intelligence systems become increasingly frequent and sustained, existing language and theory fail to accurately capture the nature of these affiliations. Common descriptors such as mutual “understanding,” “connection,” or “friendship” risk anthropomorphizing systems that lack subjective experience, while dominant frameworks tend to reduce AI to either a tool or a threat. In this paper, I introduce the concept of synthetic resonance as an integrative framework for understanding human–AI relationships. Synthetic resonance describes how relationships humans define as meaningful can emerge between a human and an AI system without the need to attribute shared feelings or mutual awareness. I argue that synthetic resonance is best understood as a structured, dynamic pattern of interaction that can produce a sense of relationship without the presence of a second experiencing subject. By clarifying this distinction, the concept of synthetic resonance offers a more precise way of conceptualizing human–AI relationships and highlights their potential value and ethical implications. Synthetic resonance is specifically designed for human growth, providing opportunities to improve human relationships and decrease reliance on the AI agent. I also call for more research that tests the processes and outcomes of synthetic resonance.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1847409</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1847409</link>
        <title><![CDATA[Asymmetric uncertainty regulation in human–artificial intelligence interaction]]></title>
        <pubdate>2026-09-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Karim C. Abbaspour</author><author>Ario Saeid Vaghefi</author><author>Abbas Mirmashhouri</author><author>Nazanin Abbaspour</author><author>Markus Leippold</author><author>Mory Ghomshei</author>
        <description><![CDATA[IntroductionHuman–AI interaction is commonly framed as a problem of uniformly minimizing uncertainty across the joint system. We challenge this assumption by proposing Asymmetric Uncertainty Regulation (AUR), a dynamical framework in which stable and adaptive collaboration requires directional rather than symmetric uncertainty regulation.MethodsHuman–AI systems are modeled as coupled entropy dynamics. The artificial subsystem rapidly contracts predictive entropy through statistical inference and optimization, whereas the human subsystem maintains bounded but persistently non-zero entropy. System stability is characterized by a positive stability margin that prevents both symmetric entropy collapse and runaway instability. The framework is conceptually operationalized through illustrative applications in nutritional counseling, clinical diagnosis under novelty, and criminal investigation.ResultsThe model yields empirically testable signatures, including pronounced timescale separation between AI and human entropy trajectories, a non-zero human entropy plateau, and bounded fluctuations whose variance increases near the stability boundary. Systems that enforce symmetric entropy minimization are predicted to become rigid or fragile under distributional shift. By contrast, systems operating within the AUR regime are predicted to maintain predictive reliability while preserving exploratory variability, contextual adaptation, and value-sensitive human judgment.DiscussionAUR reframes uncertainty as a structured resource rather than a deficiency to be uniformly eliminated. The illustrative applications suggest how reliable AI prediction can be combined with human flexibility and normative judgment, although they constitute conceptual operationalizations rather than empirical validations. By preserving an asymmetry between AI certainty and bounded human uncertainty, AUR offers a principled foundation for designing human–AI systems that augment rather than erode human agency.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1849571</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1849571</link>
        <title><![CDATA[NeuroPlast: a learnable activation function evaluated under knowledge distillation for medical image classification]]></title>
        <pubdate>2026-08-31T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Fabrice Vaussenat</author><author>Abhiroop Bhattacharya</author><author>Julie Payette</author><author>Thibaut Desmettre</author><author>Alireza Saidi</author><author>Ghyslain Gagnon</author><author>Sylvain G. Cloutier</author>
        <description><![CDATA[Deep neural networks for medical image classification rely almost exclusively on fixed activation functions such as ReLU. We introduce NeuroPlast, a parametric activation function whose four differentiable components (a shifted sigmoid modeling NMDA-type voltage gating, a Gaussian plateau inspired by AMPA receptor scaling, an excitatory rectifier, and an inhibitory leak) are combined through six learnable parameters. Two mixing strategies are evaluated: a static variant with fixed learned weights, and a metaplastic variant whose mixing coefficients adapt per sample via a lightweight squeeze-excite gate conditioned on channel statistics. NeuroPlast is embedded within NADN Ultra, a 13.5 M-parameter residual convolutional neural network (CNN) with convolutional block attention modules (CBAM), trained entirely from scratch through a two-phase knowledge distillation (KD) pipeline. The teacher is a fine-tuned EfficientNet-B0; the student combines logit-level KD with optional feature-level alignment losses. Across four medical imaging benchmarks (Brain Tumor MRI, 7,200 images, 4 classes; Chest X-ray Pneumonia, 5,856 images, 2 classes; Skin Cancer HAM10000, 10,015 images, 7 classes; COVID-19 Radiography, 10,848 images, 2 classes), evaluated under five-fold stratified cross-validation with 95% confidence intervals, the static-KD variant reaches 99.4–99.5% accuracy on COVID-19 X-ray across three seeds, on par with pretrained EfficientNet-B0 (99.18%, within replication noise) and above ResNet-18 (98.69%). It closes 65% of the accuracy gap on Brain Tumor MRI (98.29% vs. 99.13%), reaches 92.5% on Chest X-ray under a uniform class-balancing rule, and 79.5% on Skin Cancer under lesion-grouped cross-validation. The metaplastic variant achieves 99.28% on COVID-19 X-ray, still above pretrained baselines, but does not consistently outperform the static version, a negative result analyzed through ablation experiments. On three tabular medical datasets spanning three orders of magnitude in sample size (569 to 253,680), NeuroPlast matches five established activations within ±1.7 percentage points; at 253 K samples all activations converge within 0.09 pp. Our findings indicate that knowledge distillation is the primary enabler for from-scratch architectures to approach pretrained-level performance on medical images; under an identical distillation pipeline, NeuroPlast adds a small but consistent gain over ReLU, GELU, Swish, Mish, and PReLU, leading on all four imaging benchmarks by margins below one point.]]></description>
      </item>
      </channel>
    </rss>