We propose a file system with novel snapshot function, called YuihaFS. The proposed file system reduces the disk usage for the differential data by allowing users and applications to select a file for creating snapshots. With this new property, YuihaFS can reduce the amount of differential data for snapshots by avoiding creating unnecessary snapshots.
This study proposes a data expansion method to classify the excitement of conversation. This study incorporates human intuitions for conversation excitement into data augmentation. Quantification of the human intuitions would efficiently assign correct labels to the data set generated by data augmentation. The pandemic of the new coronavirus has resulted in a loss of communication opportunities. We have lost opportunities that are important to form good relationships. A deep learning model to discriminate conversation excitement would contribute to increasing such important opportunities. However, training and using models to solve real-world problems requires a lot of data. There are many cases where sufficient data cannot be collected to train a model. In such cases, data augmentation is the most promising solution. We should pay attention to the point that effective data augmentation methods vary depending on the type and characteristics of the data. This study experimentally collects conversational data. It performs data augmentation on the conversational data. It creates datasets by similarity and trains multiple models. Comparing the accuracy of these models verifies the effectiveness of incorporating human intuitions into data augmentation. The paper discusses what kind of data augmentation technique works well to generate realistic conversation data with augmentation.
Introduced a new novel weighted Grad-CAM to give more insights into how the CNN Models give their result, based on the HAM10000 dataset for skin cancer classification and Interpretability.
The study integrates a thorough investigation of machine learning approaches with unsupervised clustering algorithms to advance the area of flight phase identification in aviation. We initiate our research with a comprehensive assessment of the literature on machine learning applications for flight phase detection, both scientometrically and conceptually. This review looks at how machine learning techniques have developed to meet theoretical and practical goals in this field, as well as publishing trends, relevant papers and top research institutes.
A significant contribution of our work is the application of several unsupervised clustering algorithms to flight data, which is simulated using Monte Carlo Simulation. This allows us to identify crucial flying phases, namely the straight and level phase and the turn phase. Through the performance evaluation of these techniques, we find that among the models examined, the Gaussian Mixture Model (GMM) provides the most accurate phase detection. This result enhances the accuracy of flight phase identification and demonstrates the advantage of GMM in handling the intrinsic complexity of flight data.
Furthermore, our analysis points out areas of current research deficiency and proposes future directions for investigation, offering a roadmap for the development of machine learning methods in aviation that could improve operational effectiveness, safety and training programs.
Abstract— Urban retrofitting has emerged as a key strategy in the transition towards sustainable cities, with Positive Energy Districts (PEDs) serving as a model for achieving energy-positive urban environments. This paper explores the potential for urban retrofitting to achieve a positive energy balance through a case study of four existing districts in European Municipalities: Settimo Torinese (Italy), Großschönau (Austria), Amsterdam (Netherlands), and Resita (Romania). The analysis leverages energy balance simulations, considering various retrofitting scenarios, including building insulation, photovoltaic (PV) installations, and the adoption of flexible grid usage. The findings indicate that while achieving a PED is challenging, it is attainable through a combination of aggressive retrofitting measures, renewable energy integration, and smart energy management. The study highlights the importance of context-specific strategies, as climatic and urban characteristics significantly influence the outcomes. It aims to add to the ongoing discourse on sustainable urban development by providing empirical insights into the pathways and challenges of achieving PEDs through urban retrofitting.
Keywords — Positive Energy Districts, PED, retrofitting, Climate-Neutral Districts
In our previous study, an active power control method called variable gain control (VGC) has been proposed for grid-forming (GFM) converters. In the present paper, a reactive power control method for GFM converters to prevent an overcurrent and an alleviate a voltage oscillation under a three-line-to-ground (3LG) fault. In the proposed method, the magnitude of the voltage output of GFM converters is varied in accordance with the terminal voltage. In addition, a novel idea of variable gain and a voltage limiter is applied to avoid overcurrent. Numerical simulations demonstrate the effectiveness of the proposed method in suppressing overcurrent and voltage variation during 3LG faults.
This paper introduces a novel Dynamic Attention Layer (DAL) that enhances the generalization capabilities of Convolutional Neural Networks in both in-distribution and out-of-distribution scenarios. By dynamically adjusting attention weights based on selected percentiles during training, DAL improves the network’s ability to capture both dominant and subtle features, resulting in better accuracy and robustness across diverse datasets. The study demonstrates DAL’s effectiveness through rigorous testing, showing it outperforms traditional attention mechanisms and data augmentation techniques, offering a valuable advancement in computer vision.
This research can enhance the level of understanding of local governments in Western Australia on possible approaches to meet their net zero emissions targets.
In the field of cancer diagnosis, especially detection of brain tumors, achieving highly accurate detection is very im- portant. Deep learning, with its remarkable capabilities in object detection, has emerged as a valuable tool for identifying brain tumors. We introduce a novel approach called ViRCNN that combines the strengths of Faster R-CNN and Vision Transformer (ViT), referred to as ViRCNN. This method enhances both the accuracy and efficiency of brain tumor detection in magnetic resonance image (MRI) images. To evaluate the effectiveness of ViRCNN, we employed the Br35H dataset, which includes 801 MRI images for training, validation, and testing. Our approach demonstrates significant improvements in the Mean Average Precision 50 (MAP50) and Recall metrics compared to previous methods. Notably, ViRCNN achieves a 0.9% improvement in the MAP50 score while maintaining a parameter count of only 19 million, substantially lower than the over 80 million parameters typical of state-of-the-art methods.
This study provides a comprehensive temporal and geospatial analysis of PM2.5 levels across Asian countries from 2018 to 2023. Utilizing time series modeling (ARIMA) for predicting future pollution trends and evaluating multiple metrics for model performance, the research highlights significant regional variations in air quality and identifies key patterns in pollution trends. By categorizing countries into different pollution level clusters, the study presents a nuanced understanding of air quality dynamics, facilitating targeted policy recommendations for environmental management and public health interventions in Asia. This work contributes to the field by combining predictive modeling with spatial analysis to address a critical environmental issue.