查看更多>>摘要：The purpose of link prediction is to predict missing edges of the knowledge graph, or edges that may appear in the future. An abundance of data-driven methods exist, which rely on many labels and only consider the structural information about the graph. To address these problems, we have developed a logical-default attention graph convolution neural network model (DA-GCN). The model consists of three parts. First, the confidence of first-order logic inference with Default (DFLIN) was calculated by defining rules. Second, a local relational attention model was defined to represent self information, which includes first order, second order neighbor node information, and to train a graph convolution neural network based on iterative attention (IAGCN) by using centralized training distributed execution model. Third, iteratively update DFLIN and IAGCN using the Wasserstein distance to express the relation between entities. Finally, we conduct comparative experiments between the DA-GCN model and other models on a self-built dataset and three public data sets. The results show that the Mean Reciprocal Rank (MRR) value is about 36.48% higher than that of the R-GCN model. Our results show that DA-GCN achieves state-of-the-art performance compared to other specific or general integrated methods, especially on smallscale organized knowledge graphs.(c) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Transfer learning aims to improve the learning of the target domain with the help of knowledge from the source domain. Recently, learning using privileged information (LUPI) has been proposed to learn an accurate classifier with privileged information which is only obtained in the training stage. In this paper, we propose a new AdaBoost-based Transfer Learning with Privileged Information (AdaTLPI) method to solve the transfer learning problem with privileged information, in which AdaBoost is taken into account to combine the weak classifiers into a strong classifier. In the model, we utilize shared parameter to transfer knowledge from the source domain to the target domain. We then incorporate privileged information about the source and target domains into a unified model and AdaBoost is used to learn a strong classifier by combining the obtained weak classifiers. Finally, we present an effective optimization algorithm to solve the proposed objective model and present the boundary of training error of the proposed method. The experimental results manifest that the proposed method can outperform the previous methods.(c) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Currently, more and more researchers use the idea and strategy of three-way decision to guide scientific research. In this paper, we propose two comprehensive fuzzy concepts and apply them to construct a new three-way decision model. Firstly, with the TOPSIS method, this paper regards the optimal distance and the worst distance as the cost fuzzy concept and the benefit fuzzy concept, respectively. Then, based on the fuzzy concepts with opposite characteristics, two comprehensive fuzzy concepts (i.e., the comprehensive benefit fuzzy concept and the comprehensive cost fuzzy concept) are proposed, and the relationship between the decision-making regions corresponding to the two comprehensive fuzzy concepts is discussed. Further, four fuzzy neighborhood operators are used to describe the relationship between any two schemes and the relative loss function is used to compute the loss value of each scheme with different behaviors in different states. Afterwards, we introduce a new three-way decision model to solve the multi-criteria decision-making (MCDM) problem. Comparing analysis, Kendall analysis and Spearman analysis show that our method is feasible and stable. (c) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Privacy-preserving data consolidation is one of the prominent research topics in the big data era, mostly because of the need of consolidating data from various database communities, which are highly sensitive or competitive in nature. This leads to a new emerging problem namely how to accomplish data consolidation among different organization data sets or databases appropriately without revealing any competitive or sensitive information to non-owners. To address this issue, we first introduce a new encryption notion called "complementary set encryption" (CSE) to apply to privacy-preserving data consolidation. In this notion, to provide privacy protection, the data in one database will be encrypted under a set W prior to being sent to the other database, and a decryption key will be generated with another set Q. The data will be decrypted and consolidated into the other database, if and only if both sets W and Q are complementary. Here, the "complementary" means that the intersection of set W and set Q is empty and meanwhile the union of both is to be completed into a universal set exactly. We then describe two constructions of CSE under the public-key setting. Our first construction is designed under a weaker security notion - "payload hiding", which only preserves the data privacy but achieves a higher performance. The second construction is a stronger security notion, which we refer to as an "attribute hiding", which preserves both the privacy of the data and its associated set W. Finally, we provide a formal analysis to prove the security of our two constructions, followed by a theoretical performance comparison and an experimental evaluation. In particular, the first construction is of an independent interest in the context of cryptography since it was able to stand out for its efficiency as a predicate encryption scheme in the sense that it features both low communication overhead and computational costs simultaneously, and only requires a private key size of O(1), a ciphertext size of O(1), and only O(1) pairing computations. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Social Learning Particle Swarm Optimization (SL-PSO) greatly improves the optimization performance of PSO. In solving complex optimization problems, however, it still has some deficiencies, such as poor search ability and low search efficiency. Hence, an improved SL-PSO, namely, Three-Learning Strategy PSO (TLS-PSO) is proposed in this paper. Firstly, a med-point-example learning strategy and a random learning strategy are proposed to replace the imitation component and social influence component of SL-PSO to enhance the exploitation and exploration, respectively. Secondly, the two learning strategies are combined cleverly into an updating equation to balance exploration and exploitation. Finally, a worst-best example learning strategy is merged skillfully to construct TLS-PSO with hybrid learning mechanism and further enhance the search ability. The experimental results on the complex functions from CEC2013 and CEC2017 test sets indicate that TLS-PSO has better performance compared with state-of-the-art PSO variants and other algorithms. For example, TLS-PSO has an advantage over SL-PSO on 50 of the 56 functions from CEC2013, its running time is less than SL-PSO's and it has higher search efficiency. Simulation results on the 10 engineering problems also show that TLS-PSO outperforms 7 excellent algorithms, such as IUDE and iLSHADE,. It is expected to solve practical problems better. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Dealing with redundancy is one of the main challenges in frequency based data mining and itemset mining in particular. To tackle this issue in the most objective possible way, we introduce the theoretical bases of a new probabilistic concept: Mutual constrained independence (MCI). Thanks to this notion, we describe a MCI model for the frequencies of all itemsets which is the least binding in terms of model hypotheses defined by the knowledge of the frequencies of some of the itemsets. We provide a method for computing MCI models based on algebraic geometry. We establish the link between MCI models and a class of MaxEnt models which has already known to be used in pattern mining. As such, our research presents further insight on the nature of such models and an entirely novel approach for computing them. (C) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：This paper is about the diagnosability of fault patterns in timed stochastic discrete event systems. For this purpose, the diagnosability problem is formulated with labeled stochastic Petri net models and pure logical fault pattern nets. A particular composition of a labeled stochastic Petri net with a fault pattern net is proposed and is shown to characterize in an explicit way the fault patterns, including the timing and probabilistic aspects of the underlying system. Logical and probabilistic verifiers are derived, and used to establish a set of conditions to check not only the strong diagnosability property but also weaker notions of diagnosability. (C) 2022 Elsevier Inc. All rights reserved.
Samanta, SubhrajitPrakash, P. K. S.Chilukuri, Srinivas
21页
查看更多>>摘要：Multi-step ahead long term forecasting remains a pertinent challenge in time series literature due to non-stationary behaviour of real-world data. Predominantly most traditional time series models are parametric in nature and they use the predicted values to generate forecast for future time steps. This leads to error accumulation in each step of the forecasting horizon which causes increasingly poorer forecast in long-term. Other than the problem of error accumulation, most parametric algorithms also require significant preprocessing, hyper-parameter tuning, training and post-processing which can often put high computational burden on the system. Therefore, this paper proposes, Model Less Time-series Forecasting (MLTF), a non-parametric approach for forecasting which does not require any pre-processing or traditional training (i.e. Backpropagation). MLTF is a non-parametric method which uses statistical representations such as trend, linearity, entropy etc. to cluster series from a pre-defined repository and the series from same cluster are tagged as similar series. The trajectory of the target series is extracted from these similar series after applying an adaptive re-sampling technique. There is minimal training involved in MLTF, therefore this framework is computationally very efficient. The model-less nature also enables it to not suffer from error accumulation in long-horizon forecast. MLTF is validated empirically with a rich set of experiments involving M1, M3 competition dataset, Electricity, Volatility and COVID-19 data (over 4500 independent uni-variate series of different frequencies i.e. Hourly, Daily, Monthly, Quarterly and Yearly). The experiments demonstrate that, MLTF is significantly faster while being similar (or better) in terms of forecasting accuracy than the state-of-the-art DL methods and other non-parametric time series model. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：An effective method in machine learning often involves considerable experience with algorithms and domain expertise. Many existing machine learning methods highly rely on feature selection which are always domain-specific. However, the intervention by data scientists is time-consuming and labor-intensive. To meet this challenge, we propose a Feature Transferring Autonomous machine learning Pipeline (FTAP) to improve efficiency and performance. The proposed FTAP has been extensively evaluated on different modalities of data covering audios, images, and texts. Experimental results demonstrate that the proposed FTAP not only outperforms state-of-the-art methods on ESC-50 dataset with multi-class audio classification but also has good performance in distant domain transfer learning. Furthermore, FTAP outperforms TPOT, a state-of-the-art autonomous machine learning tool, on learning tasks. The quantitative and qualitative analysis proves the feasibility and robustness of the proposed FTAP. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：A (t, n)-threshold scheme with secure secret reconstruction, or a (t, n)-SSR scheme for short, is a (t, n)-threshold scheme against the outside adversary who has no valid share, but can impersonate a participant to take part in the secret reconstruction phase. We point out that previous bivariate polynomial-based (t, n)-SSR schemes, such as those of Harn et al. (Information Sciences 2020), are insecure, which is because the outside adversary may obtain the secret by solving a system of t(t+1)/2-ary linear equations. We revise Harn et al. scheme and get a secure (t, n)-SSR scheme based on a symmetric bivariate polynomial for the first time, where t <= n <= 2 t - 1. To increase the range of n for a given t, we construct a secure (t, n)-SSR scheme based on an asymmetric bivariate polynomial for the first time, where n >= t. We find that the share sizes of our schemes are the same or almost the same as other existing insecure (t, n)-SSR schemes based on bivariate polynomials. Moreover, our asymmetric bivariate polynomial-based (t, n)-SSR scheme is more easy to be constructed compared to the Chinese Remainder Theorem-based (t, n)-SSR scheme with the stringent condition on moduli, and their share sizes are almost the same. (C) 2022 Elsevier Inc. All rights reserved.
NSTL
Elsevier