Insignificant Nodes Research Articles

A two-stage VIKOR assisted multi-operator differential evolution approach for Influence Maximization in social networks

The impact of online social networking on daily life is extending beyond personal boundaries, becoming a tool for financial activities and even public well-being. Interactions among users and items, as nodes of such networks, and flows of information among them play a critical role in this regard. Identifying a small subset of such nodes with maximum spreading power that facilitates this process has attracted a lot of attention. The existing solution approaches to this NP-hard combinatorial optimization problem, known as the Influence Maximization (IM) problem, mainly suffer either from poor solution quality or large computational time. This study addresses these issues by proposing a two-stage optimization framework. First, to reduce the computation complexity, a multi-criteria decision making (MCDM) approach is employed to prune insignificant nodes from the candidate nodes. Second, to optimize the solutions, a modified differential evolution (DE) algorithm equipped with multiple search operators is proposed. The proposed algorithm uses the Expected Diffusion Value (EDV) in place of traditional time-consuming simulations to evaluate the fitness of a candidate solution. We proved that EDV is a monotone and submodular function, and showed that it can effectively and efficiently be integrated into a greedy-based algorithmic framework. We also provided the theoretical analysis and evaluated the performance of this proposed algorithm experimentally using two synthetic and eight diverse real-life networks. The experimental results show that our proposed algorithm achieves a better trade-off between solution quality and running time compared to the existing baseline algorithms.

Obtaining leaner deep neural networks for decoding brain functional connectome in a single shot

Neuroscientific knowledge points to the presence of redundancy in the correlations of the brain’s functional activity. These redundancies can be removed to mitigate the problem of overfitting when deep neural network (DNN) models are used to classify neuroimaging datasets. We propose an algorithm that removes insignificant nodes of DNNs in a layerwise manner and then adds a subset of correlated features in a single shot. When performing experiments with functional MRI datasets for classifying patients from healthy controls, we were able to obtain simpler and more generalizable DNNs. The obtained DNNs maintained a similar performance as the full network with only around 2% of the initial trainable parameters. Further, we used the trained network to identify salient brain regions and connections from functional connectome for multiple brain disorders. The identified biomarkers were found to closely correspond to previously known disease biomarkers. The proposed methods have cross-modal applications in obtaining leaner DNNs that seem to fit neuroimaging data better. The corresponding code is available at https://github.com/SCSE-Biomedical-Computing-Group/LEAN_CLIP.

Finding Key Node Sets in Complex Networks Based on Improved Discrete Fireworks Algorithm

Finding out the key node sets that affect network robustness has great practical significance for network protection and network disintegration. In this paper, the problem of finding key node sets in complex networks is defined firstly. Because it is an NP-hard combinatorial optimization problem, discrete fireworks algorithm is introduced to search the optimal solution, which is a swarm intelligence algorithm and is improved by the prior information of networks. To verify the effect of improved discrete fireworks algorithm (IDFA), experiments are carried out on various model networks and real power grid. Results show that the proposed IDFA is obviously superior to the benchmark algorithms, and networks suffer more damage when the key node sets obtained by IDFA are removed from the networks. The key node sets found by IDFA contain a large number of non-central nodes, which provides the authors a new perspective that the seemingly insignificant nodes may also have an important impact on the robustness of the network.

Skeleton-based action recognition using sparse spatio-temporal GCN with edge effective resistance

Graph convolutional neural networks have established significant success in solving various machine learning and computer vision problems. For skeleton-based action recognition, graph convolutional neural networks are the most suitable choice since human skeleton resembles to a graph. Stacking body skeletons over the length of video sequence results in a very complex spatio-temporal graph of many nodes and edges. Modeling the graph convolutional network directly with such a complex graph curtails the performance due to the redundancy of insignificant nodes and edges in the graph. Also for skeleton-based action recognition, the long-term contextual information is of central importance and many current architectures may fail to capture such contextual information. Therefore in order to alleviate these problems, we propose graph sparsification technique using edge effective resistance to better model the global context information and to eliminate redundant nodes and edges in the graph. Furthermore, we incorporate self-attention graph pooling to retain local properties and graph structures while pooling operation. To the best of our knowledge, we are the first to apply graph sparsification using edge effective resistance for skeleton-based action recognition and our proposed method is confirmed to be effective on action recognition, which achieves state-of-the-art results on publicly available datasets: UTD-MHAD, J-HMDB, NTU-RGB + D-60, NTU-RGB + D-120 and Kinetics dataset.

A Passive Full-Wave Micromodeling Circuit for Packaging and Interconnection Problems

A passive full-wave micromodeling circuit (FW-MMC) is proposed for high-speed/frequency interconnection problems. The circuit is derived from an FW generalized partial element equivalent circuit (G-PEEC) model, in which the radiation effect and other FW phenomena are fully taken into account. The proposed concise physically sensible circuit is obtained by recursively using a physics-based equivalent circuit transformation for absorbing the insignificant nodes of the G-PEEC model. To ensure the fidelity of the physical nature and the passivity of the MMC, the necessary broad sense criterion and the sufficient narrow sense criterion are systematically derived. It has been shown through numerical examples that the passivity violation of the physically sensible circuit model is very weak and can be effectively mediated. Thanks to the one order of magnitude reduction in the model order, the simulations for frequency- and time-domain system responses using the proposed MMC are three and two orders of magnitude faster, respectively, as compared with the G-PEEC model. Two practical examples for both frequency- and time- domains are given to demonstrate the high fidelity, scalability, accuracy, passivity, and efficiency of the FW-MMC.

A Passive PEEC-Based Micromodeling Circuit for High-Speed Interconnection Problems

A passive partial element equivalent circuit (PEEC)-based micromodeling circuit is proposed for time-domain simulation of a high-speed interconnection problem. This physics-based model order reduction method derives a concise and physically meaningful circuit model from the PEEC model by absorbing its insignificant nodes. To maintain high fidelity of the original electromagnetic PEEC model, the concept of pseudoinductor is introduced to the node-absorbing process. The derivation process does not involve any matrix inversion or decomposition and is highly suitable for GPU parallel computations. Passivity of the micromodeling circuit is ensured by a new passivity checking and enforcement method proposed for the first time. As the scale of the micromodeling circuit can be one order of magnitude smaller than that of the original PEEC model, the time-domain simulation can be three orders of magnitude faster. Two practical examples are given to demonstrate the high fidelity, scalability, and accuracy of the proposed micromodeling circuit, showing excellent applicability to high-speed interconnection problems.

Direct Mesh-Based Model Order Reduction of PEEC Model for Quasi-Static Circuit Problems

In this paper, a new concept of matrix-inversion-less derived physically expressive circuit method of model order reduction for large-scale high-speed/microwave circuit problems is proposed. It is the first model that is derived based on the mesh information of partial element equivalent circuit modeling. This new proposed method absorbs insignificant nodes and parallel branches while retaining the physical meaning of all the inductive and capacitive elements. This method does not involve any matrix inversions or decompositions and its computational overhead is dominated by outer products that can be simply accelerated by the massive GPU acceleration technique. The resultant order-reduced circuit is very useful for not only speeding up the analysis of a large-scale high-speed/microwave circuit but also interpreting the physical insight of the circuit layout from the circuit domain point of view. Three numerical examples are given, demonstrating the versatility, scalability, accuracy, and simplicity of the method.

DERIVING MEANINGFUL EQUIVALENT CIRCUITS FOR ELECTRICALLY SMALL MULTI-CONDUCTOR STRUCTURES

A new circuit reduction algorithm for generating physically meaningful equivalent circuits for electrically small structures is proposed in this work. It makes use of the generalized Y-to-Δ transformation as well as features unique to partial element equivalent circuits (PEECs) to perform the reduction process. For a given partial element equivalent circuit, insignificant nodes are removed one by one in a prioritized order according to both user-specified cut-off frequency and threshold value. By having the freedom of choosing these parameters, this algorithm provides users an option to make a tradeoff between accuracy and simplicity of the final reduced circuit. Since the generalized Y-to-Δ transformation can keep all mutual couplings intact, the order-reduced circuit should correctly capture all physical essences of the structure being modeled. Two examples are presented in this paper to validate the proposed algorithm. The equivalent circuits obtained can indeed reflect all essential physical features, demonstrating that the algorithm is a useful tool for designing and analyzing electrically small multi-conductor structures.

A Fast Adaptive Tunable RBF Network For Nonstationary Systems.

This paper describes a novel on-line learning approach for radial basis function (RBF) neural network. Based on an RBF network with individually tunable nodes and a fixed small model size, the weight vector is adjusted using the multi-innovation recursive least square algorithm on-line. When the residual error of the RBF network becomes large despite of the weight adaptation, an insignificant node with little contribution to the overall system is replaced by a new node. Structural parameters of the new node are optimized by proposed fast algorithms in order to significantly improve the modeling performance. The proposed scheme describes a novel, flexible, and fast way for on-line system identification problems. Simulation results show that the proposed approach can significantly outperform existing ones for nonstationary systems in particular.

Architecture Selection of ELM Networks Based on Sensitivity of Hidden Nodes

Extreme learning machine (ELM) was proposed as a new algorithm for training single-hidden layer feed-forward neural networks (SLFNs). One of the issues in EML is how to determine the architecture of SLFNs. Based on sensitivity of hidden nodes, an approach of architecture selection of ELM networks by applying a pruned method was proposed in this paper. The proposed pruning method utilizes sensitivity to measure the significance of hidden nodes. Beginning from an initial large number of hidden nodes, the insignificant nodes with lower sensitivity are then pruned. Experimental results on ten UCI data sets show that the proposed approach can obtain compact network architecture that generate comparable prediction accuracy on unseen samples.

A Pruning Method Based on Weight Variation Information for Feedforward Neural Networks

Artificial neural networks are powerful tools for many information processing tasks such as pattern recognition, data mining, optimization, and prediction. It is a significant problem to find optimal structures of artificial neural networks for drawing out their high computational performance. Downsizing of network structure is also an issue to be considered for hardware implementation of large-scale neural networks. In this study, we propose a pruning method to find a compact structure of feedforward neural networks with high generalization ability in classification problems. Our method evaluates the significance of neuron nodes using the information on weight variation in the training process and prune the insignificant nodes preferentially unless the classification accuracy is degraded. Numerical experiments with several benchmark datasets show that the proposed method is effective compared with other methods.

Power system restoration plan using the characteristics of scale‐free networks

AbstractA reliable electricity supply infrastructure is fundamental to modern living. An interruption to electricity services has such far reaching impact to our everyday lives to various industries compared to other service interruptions, thus to build in significant redundancy and various preventive measures to electric power system operation and planning is needed. Just as important as to incorporate redundancy and preventive measures however, an effective restoration plan needs to be in place since the possibility of electricity service interruption cannot be eliminated completely.In this paper, a novel power system restoration plan that utilizes the characteristic of so‐called scale‐free networks is proposed. For a scale‐free network, the importance of each node is determined based on the number of connections made to other nodes in that the nodes with many connections, thus important, are called “hubs.” A scale‐free network is a special complex network which follows a power law distribution, that is few hubs and many nodes with few connections at various system sizes. In the proposed plan the hubs are restored before other insignificant nodes in the system. It is shown that by doing so, the total restoration time can be reduced considerably, and notable improvements can be achieved with respect to the objective function in the mathematical formulation of electric power system restoration problem with a few constraints introduced in this paper. The effectiveness of proposed plan is tested with one of the planning problems such as valuation of black‐start capable generators using IEEE‐30 bus system. Copyright © 2008 John Wiley & Sons, Ltd.

Mesh-Based Video Coding For Low Bit-rate Communications

In this paper, a new method for low bit-rate content-adaptive mesh-based video coding is proposed. Intra-frame coding of this method employs feature map extraction for node distribution at specific threshold levels to achieve higher density placement of initial nodes for regions that contain high frequency features and conversely sparse placement of initial nodes for smooth regions. Insignificant nodes are largely removed using a subsequent node elimination scheme. The Hilbert scan is then applied before quantization and entropy coding to reduce amount of transmitted information. For moving images, both node position and color parameters of only a subset of nodes may change from frame to frame. It is sufficient to transmit only these changed parameters. The proposed method is well-suited for video coding at very low bit rates, as processing results demonstrate that it provides good subjective and objective image quality at a lower number of required bits.

Detail Preserving Wavelet-based Compression with Adaptive Context-based Quantisation

A novel compression method called MBWT (modified basic wavelet technique) for efficient medical image compression is presented. MBWT is based on classic wavelet compression scheme with dyadic decomposition, uniform scalar quantisation and entropy coding of wavelet coefficients. By exploiting of space-frequency data characteristics in wavelet domain, an algorithm of adaptive quantisation and separable context-based binary coding with two (bitewise and wordwise) arithmetic coders are proposed. Also the most suitable filter banks selection is included. Quantisation step size evaluation is done for each subband on the basis of two items: established image quality level and wavelet coefficients variance estimation. Additionally adaptive threshold data selection is performed to reduce of unimportant coefficients in noisy areas. 9-order noncausal context model depicts the importance of the data from spatial correlations. The significance of parent node is also taken into account as a frequency aspect of data dependence. The idea of entropy coding includes three main elements: a) the construction of a decomposition tree with significant and insignificant nodes, b) pruning the tree - modified zerotree construction with four symbols of alphabet, c) separate arithmetic coding of two distinct data streams - a set of significant coefficients and a significant root node map with the signs of significant data. The MBWT algorithm seems to be simpler than the majority of efficient wavelet compression techniques and the effectiveness of our method is competitive with the best algorithms in the literature across diverse classes of medical images. Objective and subjective image quality evaluation supports this conlusion. Significant efficiency improvement over SPIHT algorithm is irrefutable for all images tested.