Explainable artificial intelligence (XAI) is an important area in the AI community, and interpretability is crucial for building robust and trustworthy AI models. While previous work has explored model-level and instance-level explainable graph learning, there has been limited investigation into explainable graph representation learning. In this paper, we focus on representation-level explainable graph learning and ask a fundamental question: What specific information about a graph is captured in graph representations? Our approach is inspired by graph kernels, which evaluate graph similarities by counting substructures within specific graph patterns. Although the pattern counting vector can serve as an explainable representation, it has limitations such as ignoring node features and being high-dimensional. To address these limitations, we introduce a framework (PXGL-GNN) for learning and explaining graph representations through graph pattern analysis. We start by sampling graph substructures of various patterns. Then, we learn the representations of these patterns and combine them using a weighted sum, where the weights indicate the importance of each graph pattern's contribution. We also provide theoretical analyses of our methods, including robustness and generalization. In our experiments, we show how to learn and explain graph representations for real-world data using pattern analysis. Additionally, we compare our method against multiple baselines in both supervised and unsupervised learning tasks to demonstrate its effectiveness.

Current 3D model Level of Detail (LOD) methods require multiple models with varying detail levels to reduce client computational load, transmitting different models based on the user's distance to the object. However, this process consumes excessive network bandwidth and strains the client's memory and storage. To address this, we propose BS3 (Bézier Slicing for 3Ds), a middleware-enabled method that slices 3D meshes and fits the contours using Bézier curves. Acting as an intermediate layer, the BS3 middleware handles slicing, vectorization, sampling and reconstruction, allowing .bs3 files to be streamed only once and adjusted dynamically at different sampling rates. Our experiments demonstrate the efficiency and performance analysis of BS3, which shows that it can reduce network and storage burdens while keeping the display effect. We believe that BS3 will enhance 3D multimedia in the game, exhibition, digital museum, cultural heritage, metaverse, etc.

Graph representation learning aims to represent graphs as vectors that can be utilized in downstream tasks such as graph classification. In this work, we focus on learning diverse representations that can capture the graph information as much as possible. We propose quantifying graph information using graph entropy, where we define a probability distribution of a graph based on its nodes’ representations and global-graph representation. However, the computation of graph entropy is NP-hard due to the complex vertex-packing polytope involved in its definition. To address this challenge, we provide an approximation method leveraging orthonormal representations for graph entropy maximization. The proposed method is implemented via graph neural networks, resulting in informative node-level and graph-level representations. Experimental results demonstrate the effectiveness of our method in comparison to many baselines in unsupervised learning and semi-supervised learning tasks. The code of our method is available at https://github.com/MathAdventurer/GeMax.

This paper focuses on graph-level representation learning that aims to represent graphs as vectors that can be directly utilized in downstream tasks such as graph classification. We propose a novel graph-level representation learning principle called Lovász principle, which is motivated by the Lovász number in graph theory. The Lovász number of a graph is a real number that is an upper bound for graph Shannon capacity and is strongly connected with various global characteristics of the graph. Specifically, we show that the handle vector for computing the Lovász number is potentially a suitable choice for graph representation, as it captures a graph's global properties, though a direct application of the handle vector is difficult and problematic. We propose to use neural networks to address the problems and hence provide the Lovász principle. Moreover, we propose an enhanced Lovász principle that is able to exploit the subgraph Lovász numbers directly and efficiently. The experiments demonstrate that our Lovász principles achieve competitive performance compared to the baselines in unsupervised and semi-supervised graph-level representation learning tasks. The code of our Lovász principles is publicly available on GitHub.

Refinement of energy management in small- scale scenarios has been a hot research field in the past 10 years. An excellent algorithm can accurately infer the working condition and even the power consumption curve of every electrical appliance in a given period of time. Such information can not only help people to better control the working condition of the devices, but also improve the efficiency of energy use. From the perspective of algorithm research, an excellent dataset is essential to both the improvement of algorithm effect and the validation of algorithm efficiency; in the non-intrusive load research field, data is required to include as much information of the electrical devices as possible. Here we propose a new dataset, the Electrical Fingerprint Dataset (EFD). The data of the electrical fingerprint dataset is mainly collected from common household appliances and office devices. 126 common appliances are selected to form 15 data collection scenarios. Real and accurate annotation of the load decomposition results is obtained through synchronous acquisition of the current and voltage data at the bus and from every single device with a high frequency (1kHz). At the same time, information about the construction of the collection scenarios and the design and use of the hardware and software of special data-collection equipment, among others, will be introduced.

This paper replicates a snow depth estimation workflow in Geoweaver based on data from NASA National Snow and Ice Data Center (NSIDC), Airborne Snow Observatory (ASO), and Moderate Resolution Imaging Spectroradiometer (MODIS) datasets to make it more FAIRable for the community. The workflow searches and retrieves spatiotemporally coincident data from various data sources, preprocesses the data, and trains/tests it using a linear regression model and a deep learning model. The trained model will be able to estimate snow depth based on new remote sensed observations. Our experiments show that Geoweaver can significantly improve the capacity of sharing and synchronizing workflow among team members or individuals from other teams. Using the simple one-stop import/export button, scientists can share not only source code, but also all the history of activity and issues during their work so the people on the receiving end can see those struggles and avoid them.

Timely, frequent, and complete cropland soil moisture information acquired throughout the growing season is critical for agricultural policy, production, food security, and food prices. The NASA Soil Moisture Active and Passive (SMAP) mission provides a reliable data source for cropland soil moisture assessment. This paper presents Crop-CASMA - a web GIS application tool for cropland soil moisture monitoring and assessment based on SMAP data. This interactive Web service-based GIS application tool enables CONUS SMAP derived soil moisture data visualization, dissemination, and analytics. In this paper, we describe the Crop-CASMA application system architecture, the application implementation, and the data it serves. In addition, we also present a few snapshots of the Crop-CASMA data for cropland soil moisture monitoring. The release of Crop-CASMA greatly enhances the user experience and facilitates using soil moisture data products for crop condition monitoring and decision support.

Fine-grained plant classification has attracted great interest for its wide application. However, there is currently no suitable fine-grained plant dataset for deep learning. In this paper, we propose a method of constructing fine-grained plant dataset with adequate annotations. the dataset can help the methods to achieve better plant classification performance. Moreover, we develop an acquisition tool for collecting fine-grained plant dataset. The tool can mitigate the problem of collecting plant images in the fields.

Modeling, a common method in Earth science research, needs a significant amount of data for model initialization, validation, verification, and calibration. Most of those data requirements could be met by the Earth observation data and their derived products. However, currently the use of Earth observation data in modeling requires significant effort for data preparation. This paper presents a Web service based general framework for making Earth observation data easily accessible and usable by various Earth science models. The framework uses OGC and ISO geospatial standards and specifications for facilitating the interoperability between Earth observation data sources and Earth science models (ESMs), and geospatial processing modeling, web service workflow and product virtualizations for automatically producing model-specific data products. The framework has been implemented as CyberConnector, a building block of NSF EarthCube cyberinfrastructure. Case demonstration of CyberConnector with three representative ESMs shows the reduction of at least one order of magnitude in time and effort spent by modelers for data preparation.

It still remains a big challenge to accurately identify the geospatial objects with well-regulated outlines within remote sensing (RS) images such as residential buildings, factory storage buildings, highways, local roads, cars, and planes. In this paper, a novel spatial feature index, which is named regular shape similarity index (RSSI), is defined to address the challenge. It represents the ratio between the area of an object and its minimum bounding shape area. The application of RSSI in identifying objects with different shapes is discussed, and its capability is found to be a great supplement to the existing spatial feature hierarchy. An approach combining RSSI with object-based image analysis (OBIA) technology is proposed for image object extraction. A Web service for RSSI calculation is developed and integrated into a Web OBIA system. In the system, four experiments extracting factory storage buildings, residential buildings, roads, and planes, respectively, are conducted on three large-scale high-resolution RS images. In each experiment, two tests, i.e., one using traditional spatial features and the other using RSSI, are performed and compared. The results show that RSSI improves the accuracy of regular object extraction.

Visualization of geospatial data provenance aims to provide a user-friendly way for easy navigation and increased understanding of the derivation history of geoscientific results. Most existing work focuses on provenance modeling and management. This paper proposes to interleave map and faceted metadata for geospatial data provenance visualization and navigation. It shows how provenance in Web-geoprocessing workflows can be visualized at four levels: feature, dataset, service, and knowledge. A prototypical system, named GeoPWProv, is developed to demonstrate the applicability of the approach.

Web service technologies have shown great promise for rapid feature discovery from large volumes of remote sensing images. A complex feature is spatially composed of elementary features. The spatial relationships among elementary features can be used to discover complex features. This paper presents an ontology supported approach for detection task of complex features. The characteristic spatial relations for complex features are formalized and used for task modeling. Tasks are mapped to service ontologies to facilitate the creation of service chains. A prototype system is developed to demonstrate the applicability of the approach.

In a service-oriented geo scientific research environment, individual geospatial services must be chained together as Web geoprocessing workflows to solve a complex geoscientific problem. The development of Web geoprocessing workflows can be divided into three phases: process modeling, process model instantiation, workflow execution. Provenance, or called lineage, records the derivation history of a data product. This paper presents a provenance framework for Web geoprocessing workflows. Such a framework includes the provenance representation, provenance recording, provenance storage, provenance service, and provenance applications. The concept of "three levels of geospatial provenance" is used to advocate the categories of provenance at the knowledge, service, and data level. The three-level view addresses the derivation history in the three-phase development of Web geoprocessing workflows. The applications of provenance are demonstrated by allowing re-orchestration of geoprocessing workflows at different phases using different levels of provenance and creating a more flexible system for Web geoprocessing workflows.