Explainable deep learning for image-driven fire calorimetry

Applied Intelligence, Год журнала: 2023, Номер 54(1), С. 1047 - 1062

Опубликована: Дек. 29, 2023

Язык: Английский

---

DEEP RECURRENT NEURAL NETWORKS IN ENERGY DEMAND FORECASTING: A CASE STUDY OF KAZAKHSTAN'S ELECTRICAL CONSUMPTION

Scientific Journal of Astana IT University, Год журнала: 2024, Номер unknown

Опубликована: Янв. 12, 2024

The critical transformation of the energy sector demands innovative approaches to ensure reliability and efficiency systems. In this pursuit, study delved into potential Deep Recurrent Neural Networks (DRNNs) for forecasting demand, using a comprehensive dataset detailing Kazakhstan's electrical consumption over span two years. Traditional statistical models have historically played role in demand prediction, but growing intricacy landscape calls more advanced solutions. paper presented comparison DRNN with other traditional machine learning highlighted superior performance DRNNs, especially capturing complex temporal relationships. is confronting unprecedented challenges due population growth integration diverse sources, leading increased system strains. Accurate prediction essential reliability. models, though widely used, often overlook intricate variables like weather patterns factors. Through rigorous methodology, encompassing exploratory data analysis, feature engineering, hyperparameter optimization, an optimized model was developed. results demonstrated DRNN's exceptional capability processing time-series data, as evidenced by its attainment R-squared value 83.6%. Additionally, it achieved Mean Absolute Errors Root Squared less than 2%. However, there were noticeable deviations some predictions, suggesting areas refinement. This research underscores significance DRNNs highlighting their advantages while also noting need ongoing optimization. findings underscore promise robust tool, pivotal sector's future resilience efficiency.

Язык: Английский

---

Deep learning-based automatic pipeline for 3D needle localization on intra-procedural 3D MRI

International Journal of Computer Assisted Radiology and Surgery, Год журнала: 2024, Номер 19(11), С. 2227 - 2237

Опубликована: Март 23, 2024

Abstract Purpose Accurate and rapid needle localization on 3D magnetic resonance imaging (MRI) is critical for MRI-guided percutaneous interventions. The current workflow requires manual MRI, which time-consuming cumbersome. Automatic methods using 2D deep learning networks segmentation require image plane localization, while are challenged by the need sufficient training datasets. This work aimed to develop an automatic learning-based pipeline accurate in vivo intra-procedural MRI a limited dataset. Methods proposed adopted Shifted Window (Swin) Transformers employed coarse-to-fine strategy: (1) initial feature with Swin UNEt TRansfomer (UNETR); (2) generation of reformatted containing feature; (3) fine Transformer calculation tip position axis orientation. Pre-training data augmentation were performed improve network training. was evaluated via cross-validation 49 MR images from preclinical pig experiments. errors compared human intra-reader variation Wilcoxon signed rank test, p < 0.05 considered significant. Results average end-to-end computational time 6 s per volume. median Dice scores UNETR 0.80 0.93, respectively. 1.48 mm (1.09 pixels) 0.98°, Needle significantly smaller than (median 1.70 mm; 0.01). Conclusion achieved pixel-level without requiring large dataset has potential assist

Язык: Английский

---

An Automated Deep Learning Approach for Analyzing Stomatal Morphometry of Poplar Trees

Опубликована: Май 30, 2024

Язык: Английский

---

Analyzing the Impact of Geospatial Derivatives on Domain Adaptation with CycleGAN

Опубликована: Май 30, 2024

Язык: Английский

---