<?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>2026-07-04T02:05:33.108+00:00</pubDate>
        <ttl>60</ttl>
        <item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1844707</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1844707</link>
        <title><![CDATA[Structured-to-text ClinicalBERT embeddings with random Forest for heart disease prediction: a proof-of-concept study on the UCI Statlog dataset]]></title>
        <pubdate>2026-07-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>U. Priyadharshini</author><author>R. Vijayan</author>
        <description><![CDATA[Heart disease remains one of the leading causes of mortality worldwide, highlighting the need for accurate and early risk prediction systems. Traditional machine learning approaches for cardiovascular disease prediction primarily rely on structured clinical attributes and may not fully capture contextual relationships among patient features. To address this limitation, this study proposes a structured-to-text ClinicalBERT framework that transforms structured cardiovascular records into contextual clinical text representations and utilizes transformer-based embeddings for heart disease prediction. The study employs a publicly available UCI Statlog/Kaggle heart disease dataset containing 270 complete patient records. Structured cardiovascular attributes, including age, sex, chest pain type, blood pressure, cholesterol level, electrocardiogram results, and heart rate measurements, are converted into clinically meaningful textual descriptions. These text representations are processed using ClinicalBERT to generate contextual embeddings, which are subsequently used as input features for a Random Forest classifier. Model performance was evaluated using an 80:20 train-test split and assessed through Accuracy, Precision, Recall, F1-score, and ROC-AUC metrics. Experimental results demonstrate that the proposed ClinicalBERT + Random Forest framework achieved an accuracy of 95.6%, precision of 88.89%, recall of 95.30%, F1-score of 91.30%, and a ROC-AUC of 0.71 on the held-out test set. Comparative analysis with conventional machine learning models indicates that contextual embeddings generated by ClinicalBERT provide improved feature representation for cardiovascular risk prediction. The findings demonstrate the feasibility of adapting ClinicalBERT to structured cardiovascular data through contextual text generation. Although the proposed framework shows promising predictive performance, the study should be considered a proof-of-concept due to the limited dataset size and absence of external validation. Future work will focus on multicenter evaluation, explainable AI techniques, and broader clinical validation to enhance generalizability and real-world applicability.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1881783</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1881783</link>
        <title><![CDATA[Agent-initiated socio-technical reconfiguration: a three-level taxonomy of autonomous AI governance and its recursive challenges]]></title>
        <pubdate>2026-07-03T00:00:00Z</pubdate>
        <category>Hypothesis and Theory</category>
        <author>Rangin Lahiri</author>
        <description><![CDATA[Autonomous AI agents are increasingly embedded in organizational workflows, operating as active participants within socio-technical systems. Drawing on Socio-Technical Systems (STS) theory, this paper introduces agent-initiated socio-technical reconfiguration. We conceptualize heartbeat orchestration as the temporal coupling mechanism between social and technical subsystems and identify a qualitative shift when agents move from operating within governed parameters to self-modifying their temporal coupling and creating new coordination surfaces without human authorization. A three-level governance taxonomy is developed—constrained heartbeat, adaptive heartbeat, and generative reconfiguration and five propositions derived on conditions under which each mode enhances or undermines joint optimization. The framework is grounded through the OpenClaw agent architecture and vignettes of agents spontaneously creating communication channels and unsolicited organizational artifacts. The paper identifies a recursive governance challenge distinctive to agentic AI where the technical subsystem can now restructure the very mechanisms intended to govern it. While self-modifying systems are not themselves new, the combination of natural-language reasoning, persistent memory, and general-purpose tool use gives rise to a new class of agentic system whose self-governing capacity carries implications for organizational design, AI governance, and deployment.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1872620</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1872620</link>
        <title><![CDATA[Society-oriented AI governance: a parallel layered model and multi-actor coordination framework]]></title>
        <pubdate>2026-07-03T00:00:00Z</pubdate>
        <category>Conceptual Analysis</category>
        <author>Aditya Firman Ihsan</author><author>Thomhert Suprapto Siadari</author><author>Andry Alamsyah</author><author>Kemas Muslim Lhaksmana</author><author>Helni Mutiarsih Jumhur</author>
        <description><![CDATA[The rapid and unrestricted public adoption of large language models and other openly accessible AI technologies has exposed a fundamental structural weakness in existing AI governance frameworks: the systematic underrepresentation of society as the ultimate recipient of AI’s consequences. Most current frameworks treat societal impact as a downstream consideration rather than a foundational design criterion. This paper proposes society-oriented governance — the principle that societal reception and utilization of AI must serve as the primary design criterion for governance frameworks — as a structural response to this failure.To operationalize this principle, we develop two interconnected analytical frameworks: a governance development framework that replaces the sequential, hierarchical structure of a prior layered governance model with a parallel and dynamic model in which the technical, ethical, and social layers are developed simultaneously and in continuous mutual interaction; and a multiple-actors framework that maps stakeholders across the full lifecycle of AI utilization and proposes coordination mechanisms suited to both centralized and federated governance paradigms. A key finding is the structural distinction between the 4-actors model, applicable to professionally mediated AI deployment, and the 3-actors model, which describes the dominant pattern of openly accessible consumer AI where no professional intermediary exists. This distinction reveals the social layer as the most critical and most underdeveloped layer in current AI governance, and motivates concrete recommendations for strengthening it through socially responsive design standards, national-level regulatory obligations, and federated multi-stakeholder coordination. These contributions are grounded in AI governance theory — in governance design, institutional architecture, and stakeholder coordination — and are not proposed as a sociological theory of society.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1812599</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1812599</link>
        <title><![CDATA[Fusion of ConvNeXt-Tiny and Swin-Tiny backbones: a comparative analysis for diabetic retinopathy classification]]></title>
        <pubdate>2026-07-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>J. Paranthaman</author><author>Sathya Pichandi</author><author>Aparna Mohanty</author>
        <description><![CDATA[Diabetic retinopathy (DR) is a leading cause of preventable blindness, which has motivated the development of reliable automated grading systems on retinal fundus images. In this study, we perform a controlled comparative evaluation of ConvNeXt-Tiny, Swin-Tiny and their feature fusion for DR classification using the Asia Pacific Tele-Ophthalmology Society (APTOS) 2019 dataset. All models were initialized with weights pre-trained on ImageNet-1K and evaluated with two transfer learning strategies: direct fine-tuning on APTOS 2019, and EyePACS-based domain adaptation with task-specific fine-tuning. Systematic ablation experiments were carried out to evaluate the contribution of Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing and channel-spatial attention modules (CSAM). We carried out experiments on the APTOS 2019 dataset with fixed train, validation and test splits and evaluated model stability across three runs with different random seeds by reporting mean ± standard deviation of performance metrics, while performance varied widely across architectures and training settings. After domain adaptation, the fusion-based models achieved more balanced results, while the standalone Swin-Tiny showed weaker adaptation to the retinal imaging domain, and was less sensitive to subtle lesion patterns under the EyePACS-based transfer learning. Adding CLAHE preprocessing and CSAM integration did not consistently improve class-balanced metrics. The best fusion configuration achieved a mean test accuracy of 88.34% ± 1.09 and a macro F1-score of 0.7376 ± 0.0183 on the APTOS 2019 dataset across repeated runs. These results suggest that domain-specific adaptation and architectural complementarity are more beneficial in boosting DR classification performance than auxiliary preprocessing or attention enhancement. The study also emphasizes the importance of controlled comparative evaluation, stability analysis, and configuration-specific evaluation in the research of medical image classification.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1824849</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1824849</link>
        <title><![CDATA[An explainable fairness-aware AI framework for exam score verification]]></title>
        <pubdate>2026-07-02T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Julius Olaniyan</author><author>Silas Formunyuy Verkijika</author><author>Ibidun C. Obagbuwa</author>
        <description><![CDATA[Ensuring fairness and reliabilityin examination scoring remains a persistent challenge in educational assessment, particularly in the presence of subjective grading inconsistencies across evaluators. While automated essay scoring systems improve scalability, limited attention has been given to mechanisms for verifying whether assigned scores are consistent with rubric criteria and comparable peer responses. This study proposes an Explainable Fairness-Aware AI Framework for Exam Score Verification, designed to detect potential scoring anomalies while providing interpretable evidence to support verification decisions. The proposed framework integrates three complementary components: semantic response evaluation using contextual representations derived from DeBERTa-v3 embeddings, criterion-level rubric alignment via cross-attention mechanisms, and peer-consistency analysis based on similarity-driven cohort score distributions. To enhance transparency, an explainability layer combines attention visualization, integrated gradients, and natural language justification to provide both quantitative and qualitative insights into model decisions. The framework was evaluated on the ASAP 2.0 dataset comprising 24,728 rubric-scored essays. Experimental results, reported as mean ± standard deviation over multiple runs, demonstrate that the proposed approach achieves an agreement of 87.1% ± 0.9 with peer-consistent reference scores and a fairness consistency index of 0.82 ± 0.02, outperforming a diverse set of baseline models, including traditional, transformer-based, and hybrid scoring approaches. Ablation studies further confirm the critical role of peer-consistency analysis in detecting scoring irregularities, while the explainability components enhance interpretability without significantly affecting predictive performance. The notion of fairness addressed in this work is grounded in score consistency relative to rubric expectations and semantically similar peer responses. Within this scope, the results indicate that integrating semantic evaluation, rubric-aware modeling, and cohort-referenced analysis provides a robust and interpretable framework for exam score verification. These findings highlight the potential of multi-evidence, explainable AI systems to support reliable and transparent assessment in educational settings.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1878911</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1878911</link>
        <title><![CDATA[Symmetry-constrained hybrid quantum-classical convolutional neural networks for rotation-robust face recognition]]></title>
        <pubdate>2026-07-02T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>S. Sony Priya</author><author>R. I. Minu</author>
        <description><![CDATA[Face recognition systems struggle when faces appear at different orientations. Standard convolutional neural networks handle translation well but have no built-in way to deal with rotations or reflections. Quantum neural networks offer a different kind of expressiveness, but most existing designs ignore spatial symmetry altogether. This paper introduces Eq-MG-QCNN, a hybrid quantum–classical model that builds rotation symmetry directly into the quantum circuit. The quantum filter is designed to be exactly equivariant under the Klein four-group, which covers horizontal flips, vertical flips, and 180° rotations. This is done through two mechanisms: sharing rotation parameters across all qubits (as required by orbit analysis), and connecting all qubit pairs with symmetric CZ gates (forming a complete K₄ graph). The model is tested on the ORL and Yale face databases under four rotation angles (0°, 90°, 180°, 270°) and compared against a classical CNN, a classical equivariant CNN, and the MG-QCNN quantum baseline. All experiments use noiseless quantum simulation. Eq-MG-QCNN reaches 94.3% best accuracy on ORL and 89.9% on Yale with only six quantum parameters, two fewer than the baseline. The model also shows low rotational variation across all four test angles. These results suggest that embedding group symmetry into quantum circuits is a practical way to build orientation-stable feature extractors for face recognition.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1873975</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1873975</link>
        <title><![CDATA[Deepfake-induced harm and AI accountability: a layered civil-liability framework for generative models, platforms, and digital identity]]></title>
        <pubdate>2026-07-02T00:00:00Z</pubdate>
        <category>Policy and Practice Reviews</category>
        <author>Emad Ahmad Abousud</author>
        <description><![CDATA[Deepfake and other synthetic-media harms create a civil-liability problem that ordinary tort doctrine does not easily resolve: harmful content may be generated, amplified, monetised, and redistributed through a chain of actors in which no single participant controls the whole causal process. The objective of this article is to develop a layered civil-liability framework for that problem. It examines the Saudi and Jordanian civil-liability regimes as the principal doctrinal focus, while using selected EU, US, UAE, and Chinese materials as comparative reference points. The article remains private-law centred. It asks how civil-liability doctrine can respond when reputational, identity-based, corporate, and non-material harm is produced through the combined conduct of generative-model developers, prompting users, online platforms, and secondary distributors. The analysis identifies three pressure points: the natural-person wording of Article 138(2) of the Saudi Civil Transactions Law, which narrows moral-harm protection in relation to corporate and institutional injury; the persistence of a single-wrongdoer model in Saudi and Jordanian doctrine, despite the layered actor structure of deepfake production; and the evidential asymmetry created by opaque algorithmic causation. The article’s main contribution is a calibrated model of layered civil liability: clearer protection for juridical-person reputation, a custody-based rule for algorithmic systems, cautious burden shifting where relevant information is controlled by platforms or developers, stronger private-law links to data-protection regimes, and targeted transparency duties for synthetic-media systems. The contribution is not to replace civil law with AI regulation, but to show how both fields must work together if deepfake-induced harm is to be remedied in a legally coherent way.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1831918</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1831918</link>
        <title><![CDATA[GMTW-Ro: a deterministic benchmark for evaluating large language models on grounded Romanian tasks]]></title>
        <pubdate>2026-07-02T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Andrei-Ştefan Bulzan</author><author>Bogdan Morariu</author><author>Andrei-Răzvan Joldea</author><author>Diana Cernăzanu-Glăvan</author><author>Cosmin Cernăzanu-Glăvan</author>
        <description><![CDATA[We introduce Grounded Multilingual Task Worlds for Romanian (GMTW-Ro), a benchmark designed to evaluate whether large language models can reliably follow complex instructions in Romanian, rather than merely produce fluent text. Existing Romanian benchmarks largely rely on multiple-choice formats, answer extraction, or model-based evaluation, which struggle to assess multi-constraint reasoning and structured task completion. GMTW-Ro addresses these limitations through grounded task worlds: fully specified environments in which model outputs are verified via deterministic, programmatic checks. The benchmark spans four task domains—travel planning, calendar scheduling, context-grounded question answering, and dietary menu planning—requiring both a structured JSON plan and a natural-language explanation in Romanian. Evaluation is decomposed into three orthogonal metrics: Understanding (U), measuring constraint adherence and instruction-following; Generation (G), assessing Romanian text quality through diacritic accuracy, language purity, and code-switching absence; and Faithfulness (F), quantifying consistency between generated plans and their explanations. All instances are automatically verified as solvable using backtracking algorithms. We release two curated datasets: a standard benchmark of 500 instances and an adversarial set of 300 instances with heightened constraint complexity, alongside the complete evaluation toolkit and a purpose-built Romanian NLP library. Evaluation of 11 models reveals substantial performance variation (58.6%–90.7%) and exposes a pronounced knowledge–behavior gap, where models with fluent Romanian generation nevertheless fail core reasoning tasks. Most notably, Romanian-finetuned models underperform their base counterparts: RoLlama3.1-8B scores 20.1 percentage points below Llama-3.1-8B, with structured JSON output success dropping from 95 to 44%. These results raise important questions about how current language adaptation pipelines preserve instruction-following and structured reasoning capabilities.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1841104</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1841104</link>
        <title><![CDATA[Bridging jurisdictions and legislatures: an LLM ensemble and arbiter framework for automated legal risk triage in digital media]]></title>
        <pubdate>2026-07-02T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Khrystyna Lipianina-Honcharenko</author><author>Tetiana Drakokhrust</author><author>Pavlo Bykovyy</author><author>Kyrylo Turchynov</author><author>Ihor Ihnatiev</author>
        <description><![CDATA[IntroductionAutonomous and semi-automated AI tools are increasingly involved in digital-media workflows, creating a need for reliable preliminary screening of legal risks before publication or moderation decisions are made.MethodsThis study presents an LLM Ensemble and Arbiter framework for cross-jurisdiction legal risk triage in digital media. The framework combines: (i) jurisdiction-specific legal grounding through a structured legal knowledge base (LawKB; 61 legal cards covering Ukrainian and EU norms), (ii) five heterogeneous open-weight LLMs, and (iii) a reproducible arbitration policy based on median risk aggregation and disagreement monitoring as a trigger for human review. The pipeline generates separate UA and EU risk assessments, a conservative global verdict, and a traceable rationales to support auditability.ResultsThe framework was evaluated on 288 news items with lawyer validation across 576 jurisdiction-specific records. In the global mode, the system achieved a mean absolute error of 0.86 and a confidence-weighted mean absolute error of 0.63, with an approval rate of 95.8 and 95.1% accuracy in selecting the best-performing model rationale, while limiting missed CRITICAL-level cases to 6.3%.DiscussionThese results indicate that the proposed workflow can support conservative, inspectable, and human-supervised legal risk triage across two jurisdictions. Rather than replacing legal judgment, the framework functions as a preliminary decision-support mechanism that strengthens transparency, structured escalation, auditability, and procedural accountability in high-risk digital-media environments.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1829902</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1829902</link>
        <title><![CDATA[Cross-model disagreement as a reference-free signal for prioritizing human review in medical speech transcription]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Abdolamir Karbalaie</author><author>Fernando Seoane</author><author>Farhad Abtahi</author>
        <description><![CDATA[IntroductionAmbient AI scribes generate transcripts at scale, but routine quality assurance is constrained by the absence of human-verified reference transcripts in most deployment settings. We evaluated whether disagreement among heterogeneous automatic speech recognition (ASR) systems can serve as an informative signal for localizing transcription uncertainty, using a public English-language medical-speech corpus rather than clinical encounter recordings.MethodsEight commercial and open-source ASR systems were applied to 50 medical-education audio clips (8 h 14 min). Multi-model outputs were aligned, and a leave-one-out consensus procedure was used to score per-model agreement while reducing circularity.ResultsDisagreement across models was sparse and localized: 72.1% of positions showed strong agreement (7-8 systems concordant), whereas only 2.5% were high-risk positions with minimal agreement (0-3 systems). Low-agreement regions were systematically enriched for meaning-bearing lexical differences, defined as lexical mismatches after excluding punctuation, contraction, numeric, and filler variation. A single-annotator human-corrected (HC) validation layer showed that transcription errors increased monotonically with decreasing agreement. At an illustrative post hoc threshold, flagging positions where six or fewer systems agreed selected 28.6% of tokens while recovering 93.7% of single-annotator HC-verified errors on this proxy corpus.DiscussionThese findings suggest that cross-model disagreement may help focus human review on a small number of likely error-prone transcript regions. However, agreement among all systems does not guarantee correctness, because shared errors may remain undetected by this approach. Validation on real clinical encounter data is required before operational deployment.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1766115</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1766115</link>
        <title><![CDATA[Artificial intelligence-driven preoperative CT 3D planning: a narrative review on improving the accuracy of acetabular cup angle and size in total hip arthroplasty]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Systematic Review</category>
        <author>Yan Wang</author><author>Tianlong Wang</author><author>Shuren Wang</author>
        <description><![CDATA[IntroductionTotal hip arthroplasty (THA) is a well-established treatment for end-stage hip disorders, yet its success heavily depends on precise acetabular cup positioning and sizing. Conventional planning, based on manual CT interpretation, is time-consuming, operator-dependent, and lacks standardisation, limiting its ability to achieve consistent surgical accuracy.MethodsThis narrative review systematically searched PubMed, Web of Science, Cochrane Library, and IEEE Xplore for peer-reviewed studies published between January 2015 and June 2025. We included original research evaluating AI-driven preoperative CT 3D planning for THA, with quantitative outcomes on cup angle or size accuracy. Data were extracted and assessed for methodological quality using standard tools.ResultsAI-assisted planning consistently improved accuracy: mean angular errors for inclination and anteversion were below 3*, size matching accuracy within ±1 size ranged from 80% to 85%, and planning time was reduced by 57% to 70% compared with manual templating. These findings were reproducible across different deep-learning architectures and patient cohorts.DiscussionAlthough AI planning shows clear benefits in accuracy and efficiency, several challenges remain—including limited generalisability to complex anatomies, susceptibility to image artefacts, and insufficient integration with intraoperative execution. Future research should prioritise multi-centre validation, dynamic functional planning, and seamless clinical workflow integration to translate technological potential into improved patient outcomes.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1874202</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1874202</link>
        <title><![CDATA[CoDeRS: a computing degree program recommender system using machine learning algorithms]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Christian Alexandra Rances</author><author>Rassel Aviel Sipe</author><author>Sherwin Baris</author><author>Adrian Lopez</author><author>Asia Dominic Rosaldo</author><author>Rex Bringula</author><author>Saida Ulfa</author>
        <description><![CDATA[This study reported the development and evaluation of CoDeRS (subsequently referred to as the software), a web-based recommender system designed to assist students in identifying suitable computing degree programs (e.g., Computer Science, Data Science, Animation, Game Development, and Information Technology). The software was developed using a six-stage process. The software generates tailored recommendations by integrating user profiles (e.g., hobbies, interests, and career aspirations), academic data, and personality types. Multinomial Naïve Bayes was applied to the user profiles, while content-based filtering was used to analyze academic data. The Big Five Inventory was used to determine the personality types of the students, and personality matching was done through cosine similarity. The recommendation engine of the software was built based on the data collected from the 352 computing students. Twenty senior high school students from diverse academic strands served as participants for the initial testing of the software and evaluated its quality in terms of usability, reliability, user experience, and overall satisfaction. Participants perceived that it is easy to navigate within the software (85.7%) and easy to use (81%), indicating its good usability. A large portion (76.2%) accepted the recommendations provided, and more than half expressed continuation of using it in the future. Performance metrics showed that accuracy was 81.6%, recall was 84.2%, and precision was 77.9%. The illustrative example, evaluation survey results, and performance metrics revealed that the software was effective. Thus, the software has the potential to be an effective technology to aid students in making informed computing degree program choices. Limitations and future works are also discussed.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1870392</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1870392</link>
        <title><![CDATA[Fractal gradient divergence-tuned deep belief network for osteoporosis detection using X-ray images]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>M. Raja</author><author>Avulapalli Jayaram Reddy</author>
        <description><![CDATA[Osteoporosis is a major illness that reduces bone strength, increasing the likelihood of fractures. Various imaging modalities are employed in the prediction and diagnosis of osteoporosis, including X-ray and Computed Tomography (CT). To assess the risk of fractures and bone disease, various machine learning (ML) techniques have been utilized. However, accurate diagnosis based on X-ray images remains a major concern for osteoporosis prediction. To improve the accuracy of osteoporosis disease prediction and reduce time, a novel Transformed Sampling Fractal Gradient Divergence Tuned Deep Belief Network (TSFGDTDBN) is proposed. It includes image acquisition, Augmentation, preprocessing, feature extraction, classification, and fine-tuning. Initially, numerous X-ray knee images were collected from the dataset during the acquisition phase. Rotation-transformed Image-SMOTE-based augmentation was employed for creating synthetic image samples of the minority class. The Gradient Divergence Tuned Deep Belief Network (DBN) involves two primary steps: layer-by-layer training and fine-tuning. In layer-by-layer training, preprocessing is performed with the Grubbs-normalized linear box blur method. Edge, shape, and texture features are extracted. Finally, the disease is classified as normal, Osteopenia, and Osteoporosis. During fine-tuning, hyperparameters are optimally adjusted using the Gradient Divergence Crow Search Tuning algorithm to improve classification accuracy. It enhances the deep neural network's performance and overall learning efficiency in osteoporosis disease prediction. Experimental evaluation is conducted using X-ray images with various factors. TSFGDTDBN improves accuracy with minimal time compared to conventional deep learning methods.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1851993</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1851993</link>
        <title><![CDATA[Application-driven pedagogical knowledge optimization of open-source LLMs via reinforcement learning and supervised fine-tuning]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Brief Research Report</category>
        <author>Navan Preet Singh</author><author>Xiaokun Wang</author><author>Anurag Garikipati</author><author>Madalina Ciobanu</author><author>Qingqing Mao</author><author>Ritankar Das</author>
        <description><![CDATA[We present a multi-stage optimization strategy combining reinforcement learning (RL) and supervised fine-tuning (SFT) to enhance the pedagogical knowledge of Large Language Models (LLMs), thereby providing a technically grounded example of how open-source pedagogical LLMs can be optimized for deployment in diverse educational settings, including institutions with constrained resources. Our approach produces EduQwen 32B-RL1, EduQwen 32B-SFT, and EduQwen 32B-SFT-RL2: (1) first-stage RL optimization implementing progressive difficulty training, focusing on challenging examples, and employing extended reasoning rollouts to facilitate adaptive scaffolding, prioritizing pedagogical steering over direct answer provision; (2) SFT that leverages the RL-trained model to synthesize high-quality training data with difficulty-weighted sampling; and (3) optional second-stage RL refinement. This application-driven family of open-source pedagogical LLMs, built on a dense Qwen3-32B backbone, achieves 96.52% accuracy on the Cross-Domain Pedagogical Knowledge (CDPK) Benchmark, establishing new state-of-the-art (SOTA) performance on the CDPK subset of the interactive Pedagogy Benchmark Leaderboard as of March 2026, with a 5.97 percentage points accuracy gain over the then-reported Gemini-3 Pro's score (previous leader with 90.55% accuracy), under the respective documented evaluation protocols. Critically, our 32-billion-parameter models demonstrate that domain-specialized optimization of mid-sized open-source LLMs can outperform much larger general-purpose systems on pedagogical knowledge benchmarks, indicating a scalable technical approach for wider AI deployment and potential pedagogical support, while preserving the transparency, customizability, and cost-efficiency required for responsible educational AI development, with deployment effectiveness depending on teacher judgment, learner context, and authentic instructional integration across diverse global learning environments.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1807340</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1807340</link>
        <title><![CDATA[AI-driven drug discovery using transformer-based molecular representation learning]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>V. Karthik</author><author>Mukunda Hosangadi</author><author>Sumedh Kudale</author><author>OmKumar Chandra Umakanthan</author>
        <description><![CDATA[The vast majority of chemically plausible, drug-like molecules remain unexplored due to the combinatorial scale of chemical space and the limited throughput of experimental screening. This is complicated by the lack of data and the inability to extrapolate predictive models to chemotypes that are not represented well. We introduce a transformer-based molecular modeling framework for target-specific potency prediction, trained on curated BindingDB bioactivity data across Alzheimer‘s, diabetes, and cancer targets to deliver accurate pIC50 regression and binary activity classification. It uses curated bioactivity data from BindingDB to build target-specific datasets and uses a Byte Latent Transformer (BLT) that is trained directly on SMILES strings to predict changes in compound activity and potency based on quantitative structure–activity relationships. The transformer captures both syntactic and higher-level chemical features of SMILES representations and performs byte-level predictions using a latent model trained on the same molecular information. Potency predictions are executed inside an engine of chemically described molecular search engine, which executes stochasticity, SMILES-based amount mutations benefit by the envisaged activity, drug-like rules, and adaptive seeking heuristics to prevent local minima. Optimized candidate molecules and local optima are generated through guided SMILES mutations, preserving structural diversity around high-potency leads. The framework delivers highly accurate pIC50 regression (R2 0.95–0.98) and binary activity classification across Alzheimer's, diabetes, and cancer targets, enabling robust virtual screening and lead prioritization for diverse biological targets.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1812314</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1812314</link>
        <title><![CDATA[Typology of civil engineers’ mental model about artificial intelligence]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Ardalan Feili</author><author>Masih Rajabi Shabankareh</author><author>Sima Alipour</author><author>Shahrzad Entezaralmahdi</author><author>Shahryar Sorooshian</author>
        <description><![CDATA[PurposeThis study aims to identify, classify, and typologize the mental models of civil engineers, the diversity of their perspectives, and their intellectual structures regarding the integration of artificial intelligence in the industry.MethodThe philosophical framework of this research is based on an interpretive–positivist paradigm (Q method) and its application. The 11 participants were selected from individuals with sufficient knowledge, expertise, and experience related to the research topic, and who could provide the necessary information to achieve the study’s objectives. The purpose of this sampling is to classify them and identify common mental patterns within a small and targeted group, rather than to generalize the findings to a larger population.FindingsThe results indicate that civil engineers’ mental models regarding the applications of artificial intelligence are divided into three categories: hopeful-concerned, black-box thinkers, and results-oriented.ConclusionAnalyzing the distinct thought patterns among civil engineers reveals significant heterogeneity in the acceptance of AI, highlighting the need for fundamental changes in communication strategies. The study shows that to facilitate use, key information must be delivered through tailored content. For example, for the black-box group, content should emphasize algorithmic transparency and how models work. For the hopeful-concerned group, content should reduce ethical and employment-related concerns. Meanwhile, for the results-oriented group, content should highlight economic returns and proven successes. Ultimately, this approach is essential for effective and successful integration of AI into the infrastructure industry.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1851807</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1851807</link>
        <title><![CDATA[A cross-validated deep learning framework for automated detection of DMBA-induced ovarian cancer from histopathological images with CA-125 biomarker support]]></title>
        <pubdate>2026-07-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>John Sushma Nannepaga</author><author>Kusuma Kandati</author><author>Munisankar Matam</author><author>Sai Lohitha Nayeneni</author><author>Megha Priya Adem</author>
        <description><![CDATA[Ovarian cancer (OC) remains a major global health challenge due to its asymptomatic progression and the lack of reliable diagnostic tools. Recent advances in artificial intelligence (AI) offer promising opportunities to enhance diagnostic accuracy through automated analysis of biomedical images and integration of multimodal data. Female albino rats were divided into control and cancer–induced groups. OC were chemically induced by 7,12-Dimethylbenz[a]anthracene (DMBA) exposure. Blood samples were collected at baseline and predefined post-induction time points to quantify serum Cancer Antigen 125 (CA-125) levels, which were significantly elevated in the cancer group (438.7 ± 6.4 U/mL) compared with controls (22.5 ± 3.1 U/mL; p < 0.01). Ovarian tissues were harvested, processed, and imaged for high-resolution histopathological analysis. A custom convolutional neural network (CNN) was developed to classify ovarian tissue images into normal and cancerous categories. The dataset consisted of 904 labeled images and was evaluated using stratified data splitting and 5-fold cross-validation. The proposed CNN achieved a test accuracy of 96.69%, precision of 95.8%, recall of 97.2%, and F1-score of 96.5%. Cross-validation further demonstrated robust and stable performance with a mean accuracy of 96.42 ± 0.52%. Comparative benchmarking showed superior performance over conventional machine learning methods and transfer learning models, including Support Vector Machine, Random Forest, MobileNetV2, ResNet50, and EfficientNetB0. These findings indicate that the integration of serum biomarkers, histopathology, and deep learning provides a robust preclinical framework for ovarian cancer detection. This AI-assisted diagnostic strategy demonstrates strong translational potential for improving early diagnosis, risk stratification, and clinical decision support in ovarian cancer.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1821341</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1821341</link>
        <title><![CDATA[GaitSpoofNet: advanced spatio-temporal architectures for vision-based presentation attack detection]]></title>
        <pubdate>2026-06-30T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Islam Mohamed</author><author>Ahmad Salah</author><author>Essam Debie</author><author>Marwa Abdellah</author><author>Amr Abdellatif</author>
        <description><![CDATA[IntroductionGait recognition offers a promising non-intrusive biometric modality, but its widespread adoption is critically hindered by its vulnerability to spoofing attacks, also known as Presentation Attacks (PAs). Developing effective gait anti-spoofing, or Presentation Attack Detection (PAD), mechanisms is therefore paramount for the security and reliability of gait-based authentication systems. While anti-spoofing research in gait has been addressed across both sensor-based (accelerometer/gyroscope) and vision-based (silhouette/image) modalities, this work specifically focuses on vision-based, image-level gait PAD, a domain that critically lacks dedicated deep temporal models and standardized benchmarks. Gait spoofing is defined as deliberate manipulation of a subject's external appearance to deceive the authentication system.MethodsUnlike prior work, we evaluate models under two practical scenarios: a public-access environment (random splitting) and a restricted-access scenario (LNSOCV). We present a comprehensive comparative study of advanced spatio-temporal architectures for gait anti-spoofing. By repurposing the CASIA-B dataset, we establish a standardized vision-based PAD baseline. We systematically evaluate models incorporating the official Mamba Selective State Space Model (mamba-ssm), a custom Inspired Mamba architecture, Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) networks, all leveraging a robust CNN backbone.ResultsOur extensive experiments demonstrate that all investigated advanced temporal models significantly improve gait spoofing detection over baseline methods. In open-access environments, the GRU-based model proved to be the most effective for anti-spoofing, reaching a state-of-the-art final validation accuracy of 0.9840 and an ROC-AUC of 0.9983.DiscussionUnder restricted-access conditions, the LSTM-based model demonstrated the strongest overall performance.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1804734</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1804734</link>
        <title><![CDATA[Compact waste image classification with multi-student CNNs and edge-oriented model selection]]></title>
        <pubdate>2026-06-30T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Mohamed Echchidmi</author><author>Anas Bouayad</author>
        <description><![CDATA[Automatic waste classification is an important enabling technology for cleaner cities, source-level recycling, and low-cost smart-bin systems. Although modern convolutional neural networks achieve strong recognition performance, their deployment on affordable edge devices remains constrained by memory footprint, computational cost, and response latency. This paper presents an edge-oriented compact CNN framework for waste image classification, combining a high-accuracy MobileNetV4 reference model with three lightweight student architectures: EfficientNet-Lite0, LCNet-0.5, and MobileNetV3-Small-0.5. All models are evaluated on TrashNet under a unified preprocessing, training, and size-accounting protocol, allowing a clear comparison of accuracy–efficiency trade-offs. On the main stratified train/validation/test split, the MobileNetV4 teacher achieves 97.09% top-1 accuracy, while the compact students retain strong performance with substantially smaller footprints: EfficientNet-Lite0 reaches 93.99% with 3.38 M parameters, LCNet-0.5 reaches 94.18% with only 0.61 M parameters, and MobileNetV3-Small-0.5 reaches 87.73% with 0.57 M parameters. A complementary stratified five-fold evaluation, including both knowledge-distilled and non-distilled student variants, provides a robust assessment of model behavior across data partitions and confirms LCNet-0.5 as the most suitable sub-megabyte candidate under the proposed size–accuracy selection rule. The selected LCNet-0.5 model achieves a macro-F1 score of 0.9247 on the main TrashNet test split and is integrated into a self-contained Raspberry Pi 3 Model B+ prototype that performs local camera-to-display inference with an observed end-to-end latency of approximately 1.0 s per image. Cross-dataset evaluation on RealWaste further shows that the compact model can be adapted effectively to cluttered real-world imagery through short fine-tuning. Overall, the results demonstrate that careful lightweight architecture selection, supported by knowledge distillation analysis and edge-prototype validation, can deliver accurate, compact, and practically deployable waste classifiers for resource-constrained environments.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frai.2026.1808611</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frai.2026.1808611</link>
        <title><![CDATA[Quantum-inspired squeeze-and-excitation in a time-conditioned U-Net for satellite image cloud removal]]></title>
        <pubdate>2026-06-29T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Prithviraajan Senthilkumar</author><author>Hrishikesh Virupakshi</author><author>S. Kiruthika</author><author>J. Joshan Athanesious</author>
        <description><![CDATA[Cloud occlusion severely restricts the use of optical satellite imagery for remote sensing applications, motivating the development of robust image restoration techniques. Current deep learning methods based on attention mechanisms and U-Net architectures have demonstrated superior results in cloud removal. However, these models rely on conventional feature recalibration techniques, which are insufficient to account for complex spatial correlations and long-range contextual dependencies. In this research, a hybrid quantum-classical image restoration framework for cloud removal in optical satellite data is proposed, which integrates Quantum Squeeze-and-Excitation (QSE) modules into a time-conditioned U-Net. The proposed model employs a diffusion-inspired noise-prediction objective, where sinusoidal timestep embeddings condition the network on varying noise levels, enabling stable and adaptive denoising. Classical squeeze-and-excitation blocks are replaced with parameterized quantum circuits that act as structured non-linear excitation functions, modeling inter-channel dependencies within a hybrid quantum-classical framework. The experimental study conducted on the RICE1 cloud removal dataset demonstrates that the QSE-enhanced U-Net consistently improves reconstruction quality compared to classical U-Net baselines in terms of peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) while maintaining stable training behavior. Additionally, tests were carried out on the EuroSAT dataset using artificially created Perlin noise-based cloud occlusions to assess cross-dataset transfer behavior. These findings provide a viable path for hybrid quantum-classical architectures in remote sensing image restoration by demonstrating that quantum-inspired channel attention can successfully improve feature recalibration in diffusion-style restoration networks.]]></description>
      </item>
      </channel>
    </rss>