ABSTRACT Knowledge graph embedding (KGE) offers a more intuitive approach to discovering potential relations between known entities. However, current models are associated with challenges such as a large number of training parameters and low training efficiency and fail to provide in‐depth analysis of the impact of embedding dimensionality on entities and relations in link prediction performance. Therefore, we investigate the impact of entity and relation embedding dimensions on their interaction and assess how these dimensions affect the performance of KGE models. Based on these insights, we propose a novel dual encoder‐decoder model, NDcRE, which includes decoders MlpD and AttnMlpD, designed to capture long‐distance interactions and improve link prediction performance with fewer parameters. Evaluated on four benchmarks, WN18RR, FB15k‐237, DB100k, and YAGO3‐10, NDcRE significantly improves model efficiency by utilizing fewer parameters and dimensions, thereby enhancing both its utility and convenience. In particular, the AttnMlpD decoder further reduces the model's training parameters, enabling it to deliver strong performance even in environments with limited computational resources.

Knowledge Graph Embedding (KGE) is a powerful technique for predicting missing links in knowledge graphs. Current mainstream research primarily focuses on Transformer-based language pre-training models and graph neural network models. However, these models suffer from several issues, including a large number of parameters, low training efficiency, and dimensional explosion, which hinder their application in large-scale knowledge graphs. This paper explores the impact of entities and relations embedding dimensions on the interaction between them. Through experimental analysis, we examine how these dimensions affect the performance of KGE models. Based on these insights, we propose a novel dual encoder-decoder model called DcKE, which is designed to capture long-distance interactions and improve link prediction performance with less parameters. We conduct an extensive experimental evaluation on four widely-used datasets: WN18RR, FB15k-237, DB100k, and YAGO3-10. The results demonstrate that DcKE, by utilizing a small number of parameters and dimensions, significantly improves model efficiency, making it a promising method for large-scale knowledge graph embedding.

Wave data mining and processing are important in ocean prediction. However, wave data often exhibits imbalance, resulting in low accuracy in predicting extreme phenomena. To alleviate this problem, we propose a hybrid feature selection method based on imbalance learning (HFS-IL) to improve accuracy of prediction models. Specifically, we first use a Long Short-Term Memory (LSTM) network to train an imbalance discriminator, which aims to classify input data into common and rare subsets. Secondly, we select the optimal feature subsets by a hybrid feature selection algorithm, which is innovatively designed by combining mutual information and forward selection. To verify the effectiveness of HFS-IL, we process an imbalanced wave dataset from ERA5 by HFS-IL and use the processed data as input for an intelligent prediction model. The experimental results demonstrate that HFS-IL can effectively alleviate the impact of data imbalance and improve the accuracy of prediction, especially at station 51000, the majority of metrics outperform GRU and OSP-FEAN.

Sea ice cannot only have a significant impact on hydrological changes, climate systems, and energy balance on Earth, but it can also directly interfere with maritime activities, posing serious obstacles to ocean economic development, polar scientific research, and other activities. In today’s world, where global warming is accelerating the melting of polar sea ice, the ability to accurately identify and classify sea ice becomes particularly important. Compared with traditional statistical methods, sea ice classification methods based on machine learning and deep learning have the advantages of fast computing speed and low resource consumption. These methods require a large amount of labeled data as a driver. However, the direct labeling cost of Synthetic Aperture Radar (SAR) sea ice image datasets is high, and previous sea ice classification methods struggle to balance data processing costs with classification accuracy. To address this problem, this paper proposes a sea ice classification method based on textural features and random forest, which is called SIC-TFRF. Specifically, we first extract the textural features based on the gray-level co-occurrence matrix (GLCM) and then use the wrapper method to filter and integrate them into the polarization features of the SAR images, guiding the training of the random forest model. Extensive experiments are conducted to verify the superiority of our proposals. Particularly, the accuracy of the binary classification of sea ice and seawater and the multi-classification of different types of sea ice and seawater can reach 96.51% and 92.78%, respectively.

As an effective knowledge organizing and modeling technique, knowledge graph has become a key topic in graph research, but the practical application of KG is limited by its incompleteness. In recent years, many knowledge graph embedding(KGE) methods for knowledge graph completion(KGC) based on graph neural networks(GNN) have been proposed. However, most GNN-based KGC models are still suffer from the encoder-decoder structure of low efficiency in aggregating neighborhood information and the difficulty of model training. This paper present an optimized model that incorporates Neighborhood information into knowledge inference, to improve the performance of KGC models based on multilayer perceptron network(MLP), which is named N-MlpE. We generate an input sequence that includes the query triplet and its neighbor entities and relationships, and then feed it to an adaptive filter module to remove useless neighbors for the inference to improve the accuracy of the inference, and reduce the computational complexity of training the model. The filtered sequence is then fed into a weight calculation module and a feature extraction module simultaneously, the former is designed based on selfattention to model the relevant rule inference, which enhances the interpretability of KGE models, and the latter is based on MLP and used to capture the long-distance interactions between triplets, which can significantly improve the accuracy of inference. Extensive experiments are conducted on two standard KG datasets WN18RR and FB15k237 to verify the effectiveness of N-MlpE, the results show that the accuracy of N-MlpE model outperforms most GNN-based models.

Dimensionality reduction is a fundamental technique to address the curse of dimensionality problem in real-world big datasets. However, most existing methods either only target raw datasets that contain explicit relationships between data points, or construct the complete neighborhood graph of the dataset by calculating pairwise similarities, and then generate contexts of data points by random walking to measure the structure of the dataset, which are computationally expensive. In this paper, we propose a fast nonlinear locality-preserving dimensionality reduction approach called FVec2vec, which extends the Skip-gram model to embedding representation of general numerical matrices. Specifically, instead of constructing neighborhood graph by calculating pairwise similarities between data points, we approximate the k-nearest neighbors (kNN) of each data point in matrices by exploring its neighbors’ neighbors first. Then, we design a novel sampling algorithm to randomly sample on the kNN to depict the structure of the dataset. Experimental results show that FVec2vec is faster than most existing methods while achieving acceptable accuracy, and the accuracy is even higher than the state-of-the-art method under certain similarity metrics.

Accurately forecasting significant wave height (SWH) is meaningful since SWH is an essential parameter in coastal and ocean engineering. In order to accurately predict SWH, we propose the OSP-FEAN method, which optimizes significant wave height prediction by feature engineering and attention network. Specifically, we conduct feature engineering by adding the first-order to twelfth-order lag variables of SWH to the input set for feature enhancement and using the random forest algorithm for feature selection. Moreover, we construct a sequence to sequence neural network. In order to improve the forecast accuracy, we add an attention mechanism based on the memory layer to this neural network. Finally, extensive experiments with observed data at different stations are conducted to verify the effectiveness of our method on 6-h, 12-h and 24-h predictions, especially the superiority in outlier prediction.

Predicting the Arctic sea ice concentration (SIC) has an essential guiding role in understanding climate change trends, resource extraction and route planning. Existing deep learning models still have the problem that it is challenging to utilize the global spatial information of SIC, and the predictions of boundary regions are not accurate enough. In this paper, we propose a new deep learning model, namely ConvLSTM-CRF, to predict the monthly sea ice concentration in the Arctic. We add a dense conditional random fields to the ConvLSTM, which further extracts global spatial information and can predict SIC more accurately. The experimental results show that compared with ConvLSTM, our model has a great improvement in the overall prediction accuracy and has more accurate predictions in the SIC boundary region, especially in the melting and freezing seasons when the SIC changes drastically. Our model also shows better prediction performance when making iterative predictions. In addition, ConvLSTM-CRF can be applied to similar time series forecasting problems, such as precipitation forecasting and snowfall forecasting.

Summary The Atmospheric General Circulation Model (AGCM) as one of the most important components of Climate System Model (CSM), has been proved to be an effective way for weather forecasting and climate prediction. Although lots of efforts have been conducted to improve the computing efficiency of AGCMs, such as exploit parallel algorithms, migrating codes, and even redesigning systems to adapt to the emerging computer architectures, it is not enough to match the real requirement, due to the limited scalability of the parallel algorithms themselves. Therefore, we design and implement a scalable parallel spectral‐based atmospheric circulation mode called PAGCM in this paper. Specifically, we first analyze the data dependencies of the dimensions in different spaces according to the calculation characteristics of spectral models, and based on which we propose a two‐dimensional decomposition algorithm in PAGCM to effectively increase the involving cores for the parallel computing, and thus reduce the overall computing time. Furthermore, to adapt to the novel data decomposition in each computing stage of dynamic framework, we propose three‐dimensional data transposition algorithms and data collection algorithms correspondingly, by considering of load balancing and communication optimization. Extensive experiments are conducted on Tianhe‐2 to validate the effectiveness and scalability of our proposals.

Solving Boolean polynomial systems as an important aspect of symbolic computation, plays a fundamental role in various real applications. Although there exist many efficient sequential algorithms for solving Boolean polynomial systems, they are inefficient or even unavailable when the problem scale becomes large, due to the computational complexity of the problem and the limited processing capability of a single node. In this paper we propose an efficient parallel characteristic set method called PBCS for solving Boolean polynomial systems under the high-performance computing environment. Specifically, PBCS takes full advantage of the state-of-the-art characteristic set method and achieves load balancing by dynamically reallocating tasks. Moreover, the performance is further improved by optimizing the parameter setting. Extensive experiments are conducted to demonstrate that PBCS is efficient and scalable for solving Boolean equations, especially for the equations rasing from stream ciphers that have block triangular structure. In addition, the algorithm has good scalability and can be extended to the size of thousands CPU cores with a stable speedup.