作者机构:
[Zhang, Miao] Cent China Normal Univ, Natl Engn Res Ctr E Learning, Wuhan 430079, Peoples R China.;[Zhang, Miao; He, Tingting; Dong, Ming; Dong, M] Cent China Normal Univ, Hubei Prov Key Lab Artificial Intelligence & Smart, Wuhan 430079, Peoples R China.;[Zhang, Miao; He, Tingting; Dong, Ming; Dong, M] Cent China Normal Univ, Natl Language Resources Monitoring & Res Ctr Netwo, Wuhan 430079, Peoples R China.;[He, Tingting; Dong, Ming; Dong, M] Cent China Normal Univ, Sch Comp, Wuhan 430079, Peoples R China.
通讯机构:
[He, TT; Dong, M ] C;Cent China Normal Univ, Hubei Prov Key Lab Artificial Intelligence & Smart, Wuhan 430079, Peoples R China.;Cent China Normal Univ, Natl Language Resources Monitoring & Res Ctr Netwo, Wuhan 430079, Peoples R China.;Cent China Normal Univ, Sch Comp, Wuhan 430079, Peoples R China.
摘要:
Commonsense question answering (CQA) requires understanding and reasoning over QA context and related commonsense knowledge, such as a structured Knowledge Graph (KG). Existing studies combine language models and graph neural networks to model inference. However, traditional knowledge graph are mostly concept-based, ignoring direct path evidence necessary for accurate reasoning. In this paper, we propose MRGNN (Meta-path Reasoning Graph Neural Network), a novel model that comprehensively captures sequential semantic information from concepts and paths. In MRGNN, meta-paths are introduced as direct inference evidence and an original graph neural network is adopted to aggregate features from both concepts and paths simultaneously. We conduct sufficient experiments on the CommonsenceQA and OpenBookQA datasets, showing the effectiveness of MRGNN. Also, we conduct further ablation experiments and explain the reasoning behavior through the case study.
摘要:
Abstract Event detection plays an essential role in the task of event extraction. It aims at identifying event trigger words in a sentence and classifying event types. Generally, multiple event types are usually well‐organized with a hierarchical structure in real‐world scenarios, and hierarchical correlations between event types can be used to enhance event detection performance. However, such kind of hierarchical information has received insufficient attention which can lead to misclassification between multiple event types. In addition, the most existing methods perform event detection in Euclidean space, which cannot adequately represent hierarchical relationships. To address these issues, we propose a novel event detection network HyperED which embeds the event context and types in Poincaré ball of hyperbolic geometry to help learn hierarchical features between events. Specifically, for the event detection context, we first leverage the pre‐trained BERT or BiLSTM in Euclidean space to learn the semantic features of ED sentences. Meanwhile, to make full use of the dependency knowledge, a GNN‐based model is applied when encoding event types to learn the correlations between events. Then we use a simple neural‐based transformation to project the embeddings into the Poincaré ball to capture hierarchical features, and a distance score in hyperbolic space is computed for prediction. The experiments on MAVEN and ACE 2005 datasets indicate the effectiveness of the HyperED model and prove the natural advantages of hyperbolic spaces in expressing hierarchies in an intuitive way.
期刊:
IEEE Transactions on Geoscience and Remote Sensing,2023年61:1-11 ISSN:0196-2892
通讯作者:
Fu, LH
作者机构:
[Fu, Lihua; Chen, Xingrong; Xu, Yuejiao; Niu, Xiao] China Univ Geosci, Sch Math & Phys, Wuhan 430074, Peoples R China.;[Zhang, Meng] Cent China Normal Univ, Sch Comp Sci, Wuhan 430079, Peoples R China.
通讯机构:
[Fu, LH ] C;China Univ Geosci, Sch Math & Phys, Wuhan 430074, Peoples R China.
关键词:
Tensors;Three-dimensional displays;Matrix decomposition;Correlation;Singular value decomposition;Frequency-domain analysis;Spectral analysis;3-D seismic data reconstruction;fully connected tensor network (FCTN);Hankel tensor;low rank
摘要:
Rank-reduction approaches assume that seismic data in the frequency–space domain is of low-rank after a specific pretransformation. The presence of noise or missing traces will increase the rank; therefore, seismic data can be denoised and recovered via rank-reduction techniques. The iterative weighted projection onto convex sets (POCS) framework can be used for noise attenuation and data reconstruction simultaneously. Multichannel singular spectrum analysis (MSSA) is a classic 3-D seismic data reconstruction algorithm that rearranges the temporal frequency slices of the data with missing traces into a block Hankel matrix and then uses randomized singular value decomposition (RSVD) to interpolate slices. To further improve the efficiency and precision of 3-D seismic data reconstruction, we introduce the fully connected tensor network (FCTN) decomposition over the Hankel tensor of the frequency slices. We show that our novel rank-reduction method estimates fewer parameters than MSSA, yielding more accurate and robust results. Moreover, FCTN decomposes a fourth-order tensor into four factor contractions, which breaks the limitations that traditional tensor decomposition methods, such as CANDECOMP/PARAFAC (CP) and Tucker decomposition, cannot establish the connections between different factors and are less effective at characterizing relationships. The newly proposed approach does not require singular value decomposition (SVD), leading to an overall reduction in computational complexity. Synthetic and field examples are used to compare the performance of our method with MSSA, and our numerical results reveal the better performance of the proposed FCTN decomposition method for seismic data with large gaps or a high missing ratio.
期刊:
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS,2023年20:1-5 ISSN:1545-598X
通讯作者:
Zhang, M.
作者机构:
[Tang, Ping; Zhang, Meng] Cent China Normal Univ, Sch Comp, Wuhan 430079, Peoples R China.;[Liu, Zhihui] China Univ Geosci, Sch Math & Phys, Wuhan 430074, Peoples R China.;[Song, Rong] Cent China Normal Univ, Sch Marxism, Wuhan 430079, Peoples R China.
通讯机构:
[Zhang, M.] C;Central China Normal University, China
摘要:
Convolutional neural networks (CNNs) have become one of the most popular tools to tackle hyperspectral image (HSI) classification tasks. However, CNN suffers from the long-range dependencies problem, which may degrade the classification performance. To address this issue, this letter proposes a transformer-based backbone network for HSI classification. The core component is a newly designed double-attention transformer encoder (DATE), which contains two self-attention modules, termed spectral attention module (SPE) and spatial attention module (SPA). SPE extracts the global dependency among spectral bands, and SPA mines the local features of spatial correlation information among pixels. The local spatial tokens and the global spectral token are fused together and updated by SPA. In this way, DATE can not only capture the global dependence among spectral bands but also extract the local spatial information, which greatly improves the classification performance. To reduce the possible information loss as the network depth increases, a new skip connection mechanism is devised for cross-layer feature fusion. Experimental results in several datasets indicate that the new algorithm holds very competitive classification performance compared to the state-of-the-art methods.
作者机构:
Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning, Central China Normal University, Wuhan, Hubei, 430079, China;School of Computer, Central China Normal University, Wuhan, Hubei, 430079, China;National Engineering Research Center for E-Learning, Central China Normal University, Wuhan, Hubei, 430079, China;National Language Resources Monitor and Research Center for Network Media, Central China Normal University, Wuhan, Hubei, 430079, China;[Zhang M.] Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning, Central China Normal University, Wuhan, Hubei, 430079, China, National Engineering Research Center for E-Learning, Central China Normal University, Wuhan, Hubei, 430079, China, National Language Resources Monitor and Research Center for Network Media, Central China Normal University, Wuhan, Hubei, 430079, China
会议名称:
5th China Conference on Knowledge Graph, and Semantic Computing, CCKS 2020
会议时间:
12 November 2020 through 15 November 2020
关键词:
Deep learning;Dialogue generation;External knowledge
作者机构:
[Fang, Wenqian; Fu, Lihua; Li, Zhiming] China Univ Geosci Wuhan, Sch Math & Phys, Wuhan 430074, Peoples R China.;[Zhang, Meng] Cent China Normal Univ, Dept Comp Sci, Wuhan 430079, Peoples R China.
通讯机构:
[Li, Zhiming] C;China Univ Geosci Wuhan, Sch Math & Phys, Wuhan 430074, Peoples R China.
关键词:
interpolation;signal processing
摘要:
Seismic data interpolation is an effective way of recovering missing traces and obtaining enough information for subsequent processing. Unlike traditional methods, deep neural network (DNN)-based methods do not need to make assumptions because they can self-learn the relationship between sampled data and complete data using large training data sets and complete the interpolation with a small computational burden. However, current DNN-based approaches only focus on reducing the difference between the recovered and original data during training, which helps to improve the quality of the reconstructed seismic data as a whole, while ignoring the characteristics of the local structure. We have developed a novel seismic U-net interpolator (SUIT) algorithm based on the framework of the U-net DNN in combination with a texture loss, rather than only optimizing for reconstruction loss. Texture loss is proposed to ensure the accuracy of local structural information, which is calculated by a pretrained texture extraction neural network. Furthermore, we use a trade-off parameter to balance the reconstruction error and texture loss, and a practical technique for selecting the associated weighting parameter. The feasibility of our method is assessed via synthetic and field data examples. Numerical tests show that SUIT is robust in noisy environments and that the trained network can reconstruct irregularly or regularly sampled seismic data. Our proposed algorithm performed better than DNN-based approaches that only use reconstruction loss and the traditional low-rank matrix fitting method.
摘要:
Seismic data are often undersampled owing to physical or financial limitations. However, complete and regularly sampled data are becoming increasingly critical in seismic processing. In this paper, we present an efficient two-dimensional (2D) seismic data reconstruction method that works on texture-based patches. It performs completion on a patch tensor, which folds texture-based patches into a tensor. Reconstruction is performed by reducing the rank using tensor completion algorithms. This approach differs from past methods, which proceed by unfolding matrices into columns and then applying common matrix completion approaches to deal with 2D seismic data reconstruction. Here, we first re-arrange the seismic data matrix into a third-order patch tensor, by stacking texture-based patches that are divided from seismic data. Then, the seismic data reconstruction problem is formulated into a low-rank tensor completion problem. This formulation avoids destroying the spatial structure, and better extracts the underlying useful information. The proposed method is efficient and gives an improved performance compared with traditional approaches. The effectiveness of our patch tensor-based framework is validated using two classical tensor completion algorithms, low-rank tensor completion (LRTC), and the parallel matrix factorization algorithm (TMac), on both synthetic and field data experiments.
关键词:
heterogeneous networks;social networks;friend recommendation;co-author recommendation;random walk with restart
摘要:
It is of great importance to recommend collaborators for scholars in academic social networks, which can benefit more scientific research results. Facing the problem of data sparsity of co-author recommendation in academic social networks, a novel recommendation algorithm named HeteroRWR (Heterogeneous Random Walk with Restart) is proposed. Different from the basic Random Walk with Restart (RWR) model which only walks in homogeneous networks, HeteroRWR implements multiple random walks in a heterogeneous network which integrates a citation network and a co-authorship network to mine the k mostly valuable co-authors for target users. By introducing the citation network, HeteroRWR algorithm can find more suitable candidate authors when the co-authorship network is extremely sparse. Candidate recommenders will not only have high topic similarities with target users, but also have good community centralities. Analyses on the convergence and time efficiency of the proposed approach are presented. Extensive experiments have been conducted on DBLP and CiteSeerX datasets. Experimental results demonstrate that HeteroRWR outperforms state-of-the-art baseline methods in terms of precision and recall rate even in the case of incorporating an incomplete citation dataset.
期刊:
Communications in Computer and Information Science,2019年1095:57-73 ISSN:1865-0929
通讯作者:
Xia, Z.
作者机构:
[Sun L.; Xia Z.] Department of Computer Science and Technology, Wuhan University of Technology, Wuhan, 430063, China;Department of Computer Science, University of Surrey, Guildford, Surrey, GU2 7XH, United Kingdom;[Chen J.] Department of Computer Science and Technology, Central China Normal University, Wuhan, 430079, China;[Zhang M.] School of Computers, Hubei University of Technology, Wuhan, 430068, China;[Ho A.] Department of Computer Science and Technology, Wuhan University of Technology, Wuhan, 430063, China, Department of Computer Science, University of Surrey, Guildford, Surrey, GU2 7XH, United Kingdom
通讯机构:
[Xia, Z.] D;Department of Computer Science and Technology, China
会议名称:
5th International Symposium on Security and Privacy in Social Networks and Big Data, SocialSec 2019
会议时间:
14 July 2019 through 17 July 2019
会议论文集名称:
Security and Privacy in Social Networks and Big Data
摘要:
In this paper, the weak derivatives (WD) criterion is introduced to solve the frequency estimation problem of multi-sinusoidal signals corrupted by noises. The problem is therefore modeled as a new least squares optimization task combined with WD. To overcome the potential basis mismatch effect caused by discretization of the frequency parameters, a modified orthogonal matching pursuit algorithm is proposed to solve the optimization problem by coupling it with a novel multi-grid dictionary training strategy. The proposed algorithm is validated on a set of simulated datasets with white noise and stationary colored noise. The comprehensive simulation studies show that the proposed algorithm can achieve more accurate and robust estimation than state-of-the-art algorithms.
摘要:
Missing traces complicate the seismic data processing and may cause difficulty in geological interpretation. We present a simple but efficient normalized Gaussian weighted filter (NGWF) method for seismic data interpolation that is suitable for reconstruction despite a large number of missing traces in the data, and has low computational complexity. The missing data are filled with locally retained pixel information via the Gaussian weight. Numerical tests show that the reconstructed result using the NGWF method is better than that using dictionary learning, total variation, partial differential equation, and economic orthogonal rank-one matrix pursuit. In addition, the proposed approach is also applied to pre-stack and post-stack seismic section, and the results indicate that the new approach is applicable to the recovery of seismic data with missing traces.
作者机构:
[Zhu, Xuan; Hsu, Ching-Fang; Zhang, Maoyuan] Cent China Normal Univ, Comp Sch, Wuhan, Peoples R China.;[Harn, Lein] Univ Missouri, Dept Comp Sci Elect Engn, Kansas City, MO 64110 USA.;[Mu, Yi] Univ Wollongong, Sch Comp & Informat Technol, Wollongong, NSW 2522, Australia.
通讯机构:
[Hsu, Ching-Fang] C;Cent China Normal Univ, Comp Sch, Wuhan, Peoples R China.
关键词:
Wireless group key transfer;Vandermonde matrix;Linear secret sharing;Computation-efficient
摘要:
Efficient key establishment is an important problem for secure group communications. The communication and storage complexity of group key establishment problem has been studied extensively. In this paper, we propose a new group key establishment protocol whose computation complexity is significantly reduced. Instead of using classic secret sharing, the protocol only employs a linear secret sharing scheme, using Vandermonde Matrix, to distribute group key efficiently. This protocol drastically reduces the computation load of each group member and maintains at least the same security degree compared to existing schemes employing traditional secret sharing. The security strength of this scheme is evaluated in detail. Such a protocol is desirable for many wireless applications where portable devices or sensors need to reduce their computation as much as possible due to battery power limitations. This protocol provides much lower computation complexity while maintaining low and balanced communication complexity and storage complexity for secure group key establishment.
摘要:
Secret sharing (SS) is one of the most important cryptographic primitives used for data outsourcing. The (t, n,) SS was introduced by Shamir and Blakley separately in 1979. The secret sharing policy of the (t, n) threshold SS is far too simple for many applications because it assumes that every shareholder has equal privilege to the secret or every shareholder is equally trusted. Ito et al. introduced the concept of a general secret sharing scheme (GSS). In a GSS, a secret is divided among a set of shareholders in such a way that any "qualified" subset of shareholders can access the secret, but any "unqualified" subset of shareholders cannot access the secret. The secret access structure of GSS is far more flexible than threshold SS. In this paper, we propose an optimized implementation of GSS. Our proposed scheme first uses Boolean logic to derive two important subsets, one is called Min which is the minimal positive access subset and the other is called Max which is the maximal negative access subset, of a given general secret sharing structure. Then, conditions of parameters of a GSS are established based on these two important subsets. Furthermore, integer linear/non-linear programming is used to optimize the size of shares of a GSS. The complexity of linear/non-linear programming is O(n), where n is the number of shares generated by the dealer. This proposed design can be applied to implement GSS based on any classical SS. However, our proposed method is limited to be applicable to some general secret sharing policies. We use two GSSs, one is based on Shamir's weighted SS (WSS) using linear polynomial and the other is based on Asmuth-Bloom's SS using Chinese Remainder Theorem (CRT), to demonstrate our design. In comparing with existing GSSs, our proposed scheme is more efficient and can be applied to all classical SSs. (C) 2016 Elsevier Inc. All rights reserved.