Big Data Research最新文献

This study presents a comprehensive evaluation of two prominent deep learning models, Vision Transformer (ViT) and VGG16, within the domain of image captioning for remote sensing data. By leveraging the BLEU score, a widely accepted metric for assessing the quality of text generated by machine learning models against a set of reference captions, this research aims to dissect and understand the capabilities and performance nuances of these models across various sample sizes: 25, 50, 75, and 100 samples. Our findings reveal that the Vision Transformer model generally outperforms the VGG16 model across all evaluated sample sizes, achieving its peak performance at 50 samples with a BLEU score of 0.5507. This performance shows that ViT benefits from its ability to capture global dependencies within the data, providing a more nuanced understanding of the images. However, the performance slightly decreases as the sample size increases beyond 50, indicating potential challenges in scalability or overfitting to the training data. Conversely, the VGG16 model shows a different performance trajectory, starting with a lower BLEU score for smaller sample sizes but demonstrating a consistent improvement as the sample size increases, culminating in its highest BLEU score of 0.4783 for 100 samples. This pattern suggests that VGG16 may require a larger dataset to adequately learn and generalize from the data, although it achieves a more modest performance ceiling compared to ViT. Through a detailed analysis of these findings, the study underscores the strengths and limitations of each model in the context of image captioning. The Vision Transformer's superior performance highlights its potential for applications requiring high accuracy in text generation from images. In contrast, the gradual improvement exhibited by VGG16 suggests its utility in scenarios where large datasets are available, and scalability is a priority. This study contributes to the ongoing discourse in the AI community regarding the selection and optimization of deep learning models for complex tasks such as image captioning, offering insights that could guide future research and application development in this field.

With the increasingly severe global carbon emissions problem and the serious threat ecosystems face, carbon neutrality has gradually attracted widespread attention. This study provides an in-depth analysis of practical cases of international carbon neutrality initiatives and relevant experiences of marine cities, focusing on the construction and implementation of a legal system for economic, ecologically coordinated compensation. To evaluate the actual effectiveness of the legal system in marine cities, this study used a multiple linear regression model, considering factors such as the strictness of the legal system, enforcement efforts, and the level of participation of local enterprises and residents. The research results indicate that carbon emissions have significantly decreased in cities where legal systems are effectively enforced, from an average of 1.5 million tons per year to 1 million tons. At the same time, the economic growth rate of these cities has also significantly improved, increasing by about 2.5 percentage points from the original annual average of 4 % to 6.5 %. The study also found that the biodiversity index of these cities increased by 15 %, far higher than the average increase of 5 % in other cities, indicating the positive role of legal systems in protecting biodiversity. The public's participation rate in environmental protection activities has also increased from 25 % to 45 %, and the growth rate of green investment has reached an average of 8 % per year, far exceeding the 3 % growth rate of other cities. In terms of the ecosystem, data shows that the distribution of the ecosystem is stable, with an average ecological index of 508, which is in a relatively ideal state. The annual average growth rate of ecosystem restoration is about 3.5 %, further proving the effectiveness of ecological protection measures. Comprehensive empirical analysis shows that implementing the new legal system effectively reduces carbon emissions, enhances biodiversity, and promotes sustainable economic development. The economic growth rate increased from an average of 4.2 % to 5.1 % per year after implementing the new legal system, fully demonstrating the important role of the economic, ecologically coordinated compensation legal system in promoting carbon neutrality goals in marine cities.

Typically, graph pattern matching is expressed in terms of subgraph isomorphism. Graph simulation and its variants were introduced to reduce the time complexity and obtain more meaningful results in big graphs. Among these models, the matching subgraphs obtained by tight simulation are more compact and topologically closer to the pattern graph than results produced by other approaches. However, the number of resulting subgraphs can be huge, overlapping each other and unequally relaxed from the pattern graph. Hence, we introduce a ranking and diversification method for tight simulation results, which allows the user to obtain the most diversified and relevant matching subgraphs. This approach exploits the weights on edges of the big graph to express the interest of the matching subgraph by tight simulation. Furthermore, we provide distributed scalable algorithms to evaluate the proposed methods based on distributed programming paradigms. The experiments on real data graphs succeed in demonstrating the effectiveness of the proposed models and the efficiency of the associated algorithms. The result diversification reached 123% within a time frame that does not exceed 40%, on average, of the duration required for tight simulation graph pattern matching.

Transfer learning has proved to be effective for building predictive models even in complex conditions with a low amount of available labeled data, by constructing a predictive model for a target domain also using the knowledge coming from a separate domain, called source domain. However, several existing transfer learning methods assume identical feature spaces between the source and the target domains. This assumption limits the possible real-world applications of such methods, since two separate, although related, domains could be described by totally different feature spaces. Heterogeneous transfer learning methods aim to overcome this limitation, but they usually i) make other assumptions on the features, such as requiring the same number of features, ii) are not generally able to distribute the workload over multiple computational nodes, iii) cannot work in the Positive-Unlabeled (PU) learning setting, which we also considered in this study, or iv) their applicability is limited to specific application domains, i.e., they are not general-purpose methods.

In this manuscript, we present a novel distributed heterogeneous transfer learning method, implemented in Apache Spark, that overcomes all the above-mentioned limitations. Specifically, it is able to work also in the PU learning setting by resorting to a clustering-based approach, and can align totally heterogeneous feature spaces, without exploiting peculiarities of specific application domains. Moreover, our distributed approach allows us to process large source and target datasets.

Our experimental evaluation was performed in three different application domains that can benefit from transfer learning approaches, namely the reconstruction of the human gene regulatory network, the prediction of cerebral stroke in hospital patients, and the prediction of customer energy consumption in power grids. The results show that the proposed approach is able to outperform 4 state-of-the-art heterogeneous transfer learning approaches and 3 baselines, and exhibits ideal performances in terms of scalability.

Point cloud maps generated via LiDAR sensors using extensive remotely sensed data are commonly used by autonomous vehicles and robots for localization and navigation. However, dynamic objects contained in point cloud maps not only downgrade localization accuracy and navigation performance but also jeopardize the map quality. In response to this challenge, we propose in this paper a novel semantic SLAM approach for dynamic scenes based on LiDAR point clouds, referred to as SD-SLAM hereafter. The main contributions of this work are in three aspects: 1) introducing a semantic SLAM framework dedicatedly for dynamic scenes based on LiDAR point clouds, 2) employing semantics and Kalman filtering to effectively differentiate between dynamic and semi-static landmarks, and 3) making full use of semi-static and pure static landmarks with semantic information in the SD-SLAM process to improve localization and mapping performance. To evaluate the proposed SD-SLAM, tests were conducted using the widely adopted KITTI odometry dataset. Results demonstrate that the proposed SD-SLAM effectively mitigates the adverse effects of dynamic objects on SLAM, improving vehicle localization and mapping performance in dynamic scenes, and simultaneously constructing a static semantic map with multiple semantic classes for enhanced environment understanding.

This study presents the implementation and evaluation of a convolutional neural network (CNN) based image segmentation model using the U-Net architecture for forest image segmentation. The proposed algorithm starts by preprocessing the datasets of satellite images and corresponding masks from a repository source. Data preprocessing involves resizing, normalizing, and splitting the images and masks into training and testing datasets. The U-Net model architecture, comprising encoder and decoder parts with skip connections, is defined and compiled with binary cross-entropy loss and Adam optimizer. Training includes early stopping and checkpoint saving mechanisms to prevent overfitting and retain the best model weights. Evaluation metrics such as Intersection over Union (IoU), Dice coefficient, pixel accuracy, precision, recall, specificity, and F1-score are computed to assess the model's performance. Visualization of results includes comparing predicted segmentation masks with ground truth masks for qualitative analysis. The study emphasizes the importance of training data size in achieving accurate segmentation models and highlights the potential of U-Net architecture for forest image segmentation tasks.

This paper considers the non pilot data-aided estimation of the carrier frequency offset (CFO) and sample frequency offset (SFO) of orthogonal frequency division multiplexing (OFDM) signals in fast time-varying channel. The main obstacle is the time-variant channel response, which deteriorates the estimation validity. A practical approach to mitigate this impact is to reduce the time consumption of one-shot estimation. In this way, we propose a method to reduce the time consumption to within one OFDM symbol duration. The maximum likelihood (ML) estimator is derived based on the observations of frequency domain constellations output of two FFTs on one symbol; its closed-form approximation is then derived to reduce the calculation burden. Remarkably, our method does not require any training symbol or pilot tone embedded in the signal spectrum, therefore achieves the highest spectral efficiency. Theoretical analysis and simulation results are employed to assess the performance of proposed method in comparison with existing alternatives.

How to make a precise similarity measurement for graph data is considered as highly recommended research in many fields. Hereinto, the so-named graph data is the coalition of patterns and edges that connect patterns. By taking both of pattern information and edge information into consideration, this paper introduces an improved centrality and geometric perspective-based approach to measure the similarity between any two graph data. Once these two graph data are projected into a plane, the pattern distance can be calculated by Euclid metric. With the help of the area composed by length of each edge and angle that constructed by the positive X-axis and the edge, the area-based edge distance is computed. To get better measurement, position-based edge distance is used to modify the edge distance. Up to now, the global distance between any two graph data can be determined by combining the above mentioned two distance results. Finally, the letter dataset is applied for experiment to examine the proposed similarity approach. The experimental results show that the proposed approach captures the similarity of graph data commendably and gets a tradeoff between time and precision.

The sea surface temperature (SST) is considered an important measure for detecting changes in climate and marine ecosystems. So, its forecasting is essential for supporting governmental strategies to avoid side effects on the global population. In this paper, we analyze the SST time series and suggest that a combination between a linear component and a nonlinear component with long-term dependency can better represent it. Based on this assumption, we propose a deep neural network architecture with dilation-erosion-linear (DEL) processing units to deal with this particular kind of time series. An empirical analysis is performed in this work using three SST time series, where we explore three statistical measures. The experimental results demonstrate that the proposed model outperformed recent and classical literature forecasting techniques according to well-known performance metrics.