Sungpil Woo

This paper presents a computational pipeline for window-level semantic enrichment of texture-mapped building models used in urban digital twins (UDTs). The pipeline combines SAM-based candidate generation, super-resolution-based input matching (SRIM), a fine-tuned ResNet-50 window/non-window classifier, and texture-to-mesh mapping to instantiate verified regions as independent 3D window objects. Rather than proposing a new segmentation or super-resolution model, the study integrates existing components for low-resolution facade textures attached to 3D models. In experiments on 130 real building models, the EDSR-based SRIM configuration achieved the best mean accuracy of 95.0% and F1 score of 0.951 over 10 runs. An auxiliary experiment on cropped Open Images samples showed a consistent advantage of SRIM-based conditioning. A small-scale aspect-ratio-based evaluation of the generated 3D windows yielded an overall mean relative error of 14.93%, indicating suitability for semantic enrichment rather than precision-grade reconstruction. The method is relevant to downstream UDT applications such as facade editing, maintenance planning, and simulation-oriented model refinement.

Treeensembles usually evaluate every tree or boosting stage for every input, even when the partial prediction has already stabilized. We study inference-time lazy evaluation for pre-trained random forests and gradient-boosted decision trees. LazyRF stops random-forest evaluation using a Bayesian vote-stability score for the current leading class. LazyGBM uses a deterministic residual-capacity certificate: if the current class separation exceeds twice the maximum possible contribution of later stages, the final predicted class is guaranteed unchanged. At the default LazyRF threshold, LazyRF reduces evaluated trees by 60–90% across four benchmarks and 30 seeds while passing a paired-seed equivalence check with a 0.002 accuracy margin against full random-forest evaluation. For LazyGBM, the certified rule is conservative: certificate-only savings are 1–12% in the reported ablations, while more aggressive heuristic exits are reported separately and treated as empirically audited prototypes. We, therefore, use evaluated trees or stages as the main cost measure. Runtime measurements provide Python-prototype context rather than hardware-independent latency claims.

Healthcare federated learning faces institutional diversity, data constraints, and computing resource discrepancies among other difficulties. These problems call for creative solutions for fair and effective distributed medical image analysis that improves model performance while handling the unique data features in different healthcare environments. This paper presents RAGE-FED (Resource Aware GEneration for FedErated Disease Diagnosis), a new paradigm meant to address these important problems. Emphasising computational efficiency, communication overhead reduction, and energy saving, RAGE FED seeks to maximize performance across healthcare facilities while ensuring diagnostic accuracy and privacy. To achieve these objectives the framework makes use of advanced data generation methods and adaptive training systems. To effectively use the different processing capacity of the participating nodes, it uses dynamic model complexity adjustment and adaptive learning rate scheduling. To protect private patient data, RAGE-FED additionally uses privacy-preserving techniques like differential privacy and safe aggregation. By use of synthetic data creation, the framework also solves data shortage and imbalance problems. The performance evaluation on the Chest X-Ray Pneumonia dataset confirms the effectiveness of RAGE-FED in diverse healthcare settings, showing gains of over 21.47% in efficiency, communication, and energy use compared to state-of-the-art methods. Extended experiments on additional datasets further validate generalizability, with improvements of 20.13% on Brain Tumor MRI, and 17.85% on HAM10000 skin lesion images. Together with ablation and sensitivity analyses, these results demonstrate the robustness, scalability, and broad applicability of the proposed framework.

Emotion recognition based on electroencephalography (EEG) signals has many potential applications. However, the development of an emotion recognition model for real-world deployment encounters significant setbacks in learning an invariant representation due to data distribution shift. Some prior studies adopted domain adaptation methods to tackle this issue. However, these studies perform marginally to tackle distribution shift in the source data and neglect to focus on the extraction of domain and target-specific features, which are crucial for domain invariant representation learning. In this study, we propose an assisted domain adaptive representation learning (ADARL) method that learns domain invariant and emotion-specific features for recognizing emotions across different subjects. The proposed architecture introduces novel attention modules to segregate domain-specific and emotion-specific feature maps by exploiting the inter-relationship between different channels and bands. The domain-specific residual feature maps (peripheral features) extracted by the proposed attention mechanism are used to assist the domain discriminator that forces the model to learn a generalized representation of EEG signals. To tackle the inter-subject variability in the source data, we introduce a novel class-aware compactness loss function to improve the domain adaptation transfer efficiency. The experiments conducted on three EEG datasets (SEED, SEED-IV, and DEAP) demonstrate that the ADARL method significantly improves emotion recognition performance and efficiently alleviates the domain shift problem by learning a generalized representation of EEG signals.