Fluent AI-generated critique does not by itself establish educational validity. ArtCreaS links construct-overlap diagnostics, missing-evidence handling, and uncertainty reporting to a proposed teacher-controlled feedback workflow. The evaluated assessment core uses ExtraTrees and ridge regression; no generative model or classroom feedback intervention was evaluated. The primary benchmark comprises 3819 creativity-labeled APDDv2 records, including 1877 with comments. Across five stratified 80/20 splits, Reliability-Gated, Craft-adjusted Multimodal Fusion (RGC-MF) achieved MAE = 0.306 ± 0.007, Pearson r = 0.954, and R2 = 0.910 on the 1–10 score scale. The 0.002 MAE difference from structured ExtraTrees does not establish a meaningful or statistically supported improvement. At 50% independent rubric dropout, five-split, three-mask evaluation yielded MAE = 0.506 ± 0.009 for RGC-MF and 0.593 ± 0.011 for early fusion. A pooled out-of-fold residual interval attained mean coverage of 88.1% overall and 83.8% for sketches at a nominal 90% target. The stress tests concern controlled random evidence loss, while the fixed-width interval is a descriptive heuristic rather than standard split conformal prediction. Public AuDrA outputs provide a separate visual-scoring anchor (r = 0.802), rather than external validation of RGC-MF. The contribution is an assessment architecture supported by repeated prediction, robustness, and subgroup analyses. Generative wording, teacher release, student contestation, and learning effects remain proposed validation targets.
This research was based on teaching experience as an architecture lecturer who taught a lighting course for undergraduate students in the Department of Architecture. Simulation software is often used to teach lighting to learners. However, in practice, many undergraduate students only focus on the final visualization results without understanding it in depth. This study aims to evaluate the effectiveness of an integrated teaching approach for lighting education that combines theory, simulation, and prototyping. The learning experiences, challenges, and perceptions toward computational lighting analysis were then examined. In this study, only one lecture was conducted, and as a control, a teacher from another university provided insights regarding the fundamentals of natural lighting. Furthermore, senior students (fourth year) evaluated the work of their juniors. Objective knowledge and assignment evaluations were assessed before and after the workshop to determine the teaching framework’s impact. This showed that 72% of students successfully completed the assignments, even though the initial pre-workshop survey revealed that only 30% of students had never performed building simulation. Furthermore, 100% of the students were satisfied with the workshop, and 81% expressed a desire to learn more and use them in the workplace.
Neonatal cerebral atrophy is a serious form of structural brain anomaly observed in up to 15% of newborns admitted to neonatal intensive care units (NICUs). Early and accurate assessment is clinically important because it can inform prognosis, guide neuroprotective and rehabilitative management, and support long-term developmental monitoring. This systematic review, conducted and reported in accordance with the PRISMA 2020 statement, searched four electronic databases (PubMed/MEDLINE, Scopus, Web of Science, and IEEE Xplore) from inception to August 2026, yielding 312 records; after two-stage screening by two independent reviewers, 90 peer-reviewed studies were included and assessed for reporting quality using the CLAIM checklist. The review found that deep learning methods for neonatal brain MRI are advancing rapidly in tissue segmentation, brain age estimation, injury detection, and anomaly detection; however, no included study directly addressed the diagnosis of neonatal cerebral atrophy, and no validated atrophy-specific model, dataset, or benchmark currently exists. Transferable methodological foundations were identified in convolutional and U-Net architectures, Transformers, and Mamba-based state-space models, the latter offering linear-complexity long-range modeling that is potentially advantageous for high-resolution 3D neonatal MRI; these architectures are presented as promising research options that still require direct comparative validation. Publicly available neonatal MRI datasets lack radiologist-confirmed atrophy labels, and explainable AI methods remain largely unvalidated in neonatal neuroimaging. The review concludes by defining priority directions: neonatal-specific 3D architectures, age-normalized volumetric biomarkers, externally validated explainable pipelines, and curated atrophy-labeled datasets.
Accurately identifying whether a vehicle is controlled by a human driver or by an autonomous policy is critical in future mixed-traffic environments for safety monitoring and liability attribution. To achieve accurate classification or identification, this paper proposes NMTT-Net (Noise-agnostic Multimodal Temporal Transformer Network), a multimodal time-series classifier that combines VAE-derived latent embeddings from front-view camera images with 18-dimensional vehicle dynamics signals. To maintain reliability under adverse weather and sensor corruption, NMTT-Net applies modality-specific noise detection and denoising prior to temporal encoding, and learns discriminative behavioral patterns using dual-path Temporal Transformer encoders with fixed-weight fusion. We construct a Human-to-Autonomy (H2A) dataset in the CARLA simulator under Town07 containing Human, PPO, and DDQN trajectories under clean and noisy conditions, in which image noise is induced by adverse weather and dynamics noise is modeled as Gaussian corruption. Experimental results show that NMTT-Net achieves 99.509% and 98.617% accuracy under clean and noisy conditions, respectively, resulting in a robustness gap (clean-to-noisy accuracy drop) of 0.892%, which is smaller than that of denoised single-modality baselines. These simulation-bound results demonstrate that noise-aware multimodal temporal fusion can enable reliable H2A identification under severe environmental degradation, although further study is necessary for real-world readiness.
Commercial cloud-controlled service robots are increasingly deployed in shared public facilities, yet many remain task-centric delivery platforms that cannot be easily extended with surrounding-people recognition, situated communication policy, or continuous access to internal middleware. This article proposes the Local Safety Companion Robot Extension (LSCRx), a non-invasive local edge companion architecture that augments such robots with human-aware communication while preserving manufacturer-controlled cloud operation and built-in safety functions. The central contribution is a communication-space design method that interprets robot–user distance, distance dynamics, body cues, field-of-view availability, and recognition confidence as service-level evidence. In the proposed conceptual policy, a gradual, confidence-supported decrease in distance would be treated as participation evidence, whereas separation, stopping outside the service zone, target loss, or unstable recognition would lead to waiting, distant greeting, information display, or other distance-preserving service actions. A prototype combines a commercial service robot with smartphone-class companion modules and a local edge server to demonstrate local body tracking, self-localization, data transmission, and cloud-API command preparation. Viewpoint-aware selection among registered spatial service waypoints (RSSWs) and anti-chasing rules are proposed as a conceptual model and were not implemented or evaluated in the present prototype. In the baseline route study, offline translational correction using known route-endpoint coordinates reduced the mean return-segment human-position error from 0.770 m to 0.476 m (approximately 38.2%) on the recorded data; this reduction was not independently validated. The evaluation was limited to one robot at a single university indoor location. Exploratory experiments involved seven participants, yielding six valid body-tracking datasets and seven questionnaire responses. These experiments and field observations suggest that contactless cues, camera placement, stop distance, and field of view jointly influence service usability under the tested conditions. The findings suggest a potential HRI design gap: the distance that is socially appropriate for a person may not coincide with the distance that is technically appropriate for reliable skeleton recognition. These preliminary findings further suggest that near-distance service interaction may benefit from both distance-aware robot policy and robust wide-field smartphone-based AR/body-tracking support that can preserve skeleton completeness and metric spatial stability. Further evaluation is needed to assess generalizability beyond this setting.
