Research

To improve the performance of code readability assessment and enhance the practicality of existing assessment approaches, we propose a repository-oriented code readability assessment method based on Heterogeneous Graph Attention Network (HAN) to learn developer-code contribution relationship.

This paper presents a classification model for seismic data using Mel-spectrum analysis and pre-trained ResNet34 with transfer learning, achieving an accuracy of 98.32%. It introduces a more efficient approach for seismic wave pattern recognition in oil exploration compared to traditional Fourier transformation methods.

This paper reviews various neural network-based models for grayscale image colorization, focusing on CNN, GAN, and DNN-based techniques. It highlights the evolution of these methods and discusses their application in real-world scenarios.

This study focuses on employing machine learning techniques, including advanced deep learning models like ResNet, VGGNet, DenseNet, and a baseline SVM and custom CNN, to discern between AI-generated and genuine images. The research uses the CIFAKE dataset and aims to advance the field of digital media integrity.

This study presents an adaptive-grained method for estimating CO2 emissions using spatiotemporal cells, focusing on heterogeneous vehicles. It leverages PEMS technology to collect emissions data and applies DNN and LSTM models for carbon emission prediction in varying grid sizes, aimed at improving accuracy and adaptability in different urban scenarios.