Fully invertible hyperbolic neural networks for segmenting large-scale surface and sub-surface data

The large spatial/temporal/frequency scale of geoscience and remote-sensing datasets causes memory issues when using convolutional neural networks for(sub-)surface data segmentation.Recently developed fully reversible or fully invertible networks can mostly avoid memory limitations by recomputing the states during the backward pass through the network.This results in a low and fixed memory requirement for storing network states,as opposed to the typical linear memory growth with network depth.This work focuses on a fully invertible network based on the telegraph equation.While reversibility saves the major amount of memory used in deep networks by the data,the convolutional kernels can take up most memory if fully invertible networks contain multiple invertible pooling/coarsening layers.We address the explosion of the number of convolutional kernels by combining fully invertible networks with layers that contain the convolutional kernels in a compressed form directly.A second challenge is that invertible networks output a tensor the same size as its input.This property prevents the straightforward application of invertible networks to applications that map between different input-output dimensions,need to map to outputs with more channels than present in the input data,or desire outputs that decrease/increase the resolution compared to the input data.However,we show that by employing invertible networks in a non-standard fashion,we can still use them for these tasks.Examples in hyperspectral land-use classification,airborne geophysical surveying,and seismic imaging illustrate that we can input large data volumes in one chunk and do not need to work on small patches,use dimensionality reduction,or employ methods that classify a patch to a single central pixel.

Sewage flow optimization algorithm for large-scale urban sewer networks based on network community division

By considering the flow control of urban sewer networks to minimize the electricity consumption of pumping stations, a decomposition-coordination strategy for energy savings based on network community division is developed in this paper. A mathematical model characterizing the steady-state flow of urban sewer networks is first constructed, consisting of a set of algebraic equations with the structure transportation capacities captured as constraints. Since the sewer networks have no apparent natural hierarchical structure in general, it is very difficult to identify the clustered groups. A fast network division approach through calculating the betweenness of each edge is successfully applied to identify the groups and a sewer network with arbitrary configuration could be then decomposed into subnetworks. By integrating the coupling constraints of the subnetworks, the original problem is separated into N optimization subproblems in accordance with the network decomposition. Each subproblem is solved locally and the solutions to the subproblems are coordinated to form an appropriate global solution. Finally, an application to a specified large-scale sewer network is also investigated to demonstrate the validity of the proposed algorithm.

Spanning tree-based algorithm for hydraulic simulation of large-scale water supply networks

With the purpose of making calculation more efficient in practical hydraulic simulations, an improved algorithm was proposed and was applied in the practical water distribution field. This methodology was developed by expanding the traditional loop-equation theory through utilization of the advantages of the graph theory in efficiency. The utilization of the spanning tree technique from graph theory makes the proposed algorithm efficient in calculation and simple to use for computer coding. The algorithms for topological generation and practical implementations are presented in detail in this paper. Through the application to a practical urban system, the consumption of the CPU time and computation memory were decreased while the accuracy was greatly enhanced compared with the present existing methods.

Intelligent Networking Technology and Experimental Demonstration of Large-Scale Heterogeneous Optical Networks

A novel routing architecture named DREAMSCAPE is presented to solve the problem of path computation in multi-layer, multi-domain and multi-constraints scenarios, which includes Group Engine (GE) and Unit Engine (UE). GE, UE and their cooperation relationship form the main feature of DREAMSCAPE, i.e. Dual Routing Engine (DRE). Based on DRE, two routing schemes are proposed, which are DRE Forward Path Computation (DRE-FPC) and Hierarchical DRE Backward Recursive PCE-based Computation (HDRE-BRPC). In order to validate various intelligent networking technologies of large-scale heterogeneous optical networks, a DRE-based transport optical networks testbed is built with 1000 GMPLS-based control nodes and 5 optical transport nodes. The two proposed routing schemes, i.e. DRE-FPC and HDRE-BRPC, are validated on the testbed, compared with traditional Hierarchical Routing (HR) scheme. Experimental results show a good performance of DREAMSCAPE.

Active Queue Management within Large-Scale Wired Networks

Signal transmission control protocol sources with the objective of managing queue utilization and delay is actually a feedback control problem in active queue management (AQM) core routers. This paper extends AQM control design for single network systems to large-scale wired network systems with time delays at each communication channel. A system model consisted of several local networks is first constructed. The stability condition guaranteeing overall stability is subsequently derived using Lyapunov stability theory. The results developed have been successfully verified on a network simulator.

An Adaptive Sliding Mode Tracking Controller Using BP Neural Networks for a Class of Large-scale Nonlinear Systems

A new type controller, BP neural-networks-based sliding mode controller is developed for a class of large-scale nonlinear systems with unknown bounds of high-order interconnections in this paper. It is shown that decentralized BP neural networks are used to adaptively learn the uncertainty bounds of interconnected subsystems in the Lyapunov sense, and the outputs of the decentralized BP neural networks are then used as the parameters of the sliding mode controller to compensate for the effects of subsystems uncertainties. Using this scheme, not only strong robustness with respect to uncertainty dynamics and nonlinearities can be obtained, but also the output tracking error between the actual output of each subsystem and the corresponding desired reference output can asymptotically converge to zero. A simulation example is presented to support the validity of the proposed BP neural-networks-based sliding mode controller.

Estimation of the anisotropy of hydraulic conductivity through 3D fracture networks using the directional geological entropy

With an extension of the geological entropy concept in porous media,the approach called directional entrogram is applied to link hydraulic behavior to the anisotropy of the 3D fracture networks.A metric called directional entropic scale is used to measure the anisotropy of spatial order in different directions.Compared with the traditional connectivity indexes based on the statistics of fracture geometry,the directional entropic scale is capable to quantify the anisotropy of connectivity and hydraulic conductivity in heterogeneous 3D fracture networks.According to the numerical analysis of directional entrogram and fluid flow in a number of the 3D fracture networks,the hydraulic conductivities and entropic scales in different directions both increase with spatial order(i.e.,trace length decreasing and spacing increasing)and are independent of the dip angle.As a result,the nonlinear correlation between the hydraulic conductivities and entropic scales from different directions can be unified as quadratic polynomial function,which can shed light on the anisotropic effect of spatial order and global entropy on the heterogeneous hydraulic behaviors.

An estimation formula for the average path length of scale-free networks

A universal estimation formula for the average path length of scale free networks is given in this paper. Different from other estimation formulas, most of which use the size of network, N, as the only parameter, two parameters including N and a second parameter α are included in our formula. The parameter α is the power-law exponent, which represents the local connectivity property of a network. Because of this, the formula captures an important property that the local connectivity property at a microscopic level can determine the global connectivity of the whole network. The use of this new parameter distinguishes this approach from the other estimation formulas, and makes it a universal estimation formula, which can be applied to all types of scale-free networks. The conclusion is made that the small world feature is a derivative feature of a scale free network. If a network follows the power-law degree distribution, it must be a small world network. The power-law degree distribution property, while making the network economical, preserves the efficiency through this small world property when the network is scaled up. In other words, a real scale-free network is scaled at a relatively small cost and a relatively high efficiency, and that is the desirable result of self-organization optimization.

New results on global exponential stability of competitive neural networks with different time scales and time-varying delays

This paper studies the global exponential stability of competitive neural networks with different time scales and time-varying delays. By using the method of the proper Lyapunov functions and inequality technique, some sufficient conditions are presented for global exponential stability of delay competitive neural networks with different time scales. These conditions obtained have important leading significance in the designs and applications of global exponential stability for competitive neural networks. Finally, an example with its simulation is provided to demonstrate the usefulness of the proposed criteria.

A sub-grid scale model for Burgers turbulence based on the artificial neural network method

The present study proposes a sub-grid scale model for the one-dimensional Burgers turbulence based on the neuralnetwork and deep learning method.The filtered data of the direct numerical simulation is used to establish thetraining data set,the validation data set,and the test data set.The artificial neural network(ANN)methodand Back Propagation method are employed to train parameters in the ANN.The developed ANN is applied toconstruct the sub-grid scale model for the large eddy simulation of the Burgers turbulence in the one-dimensionalspace.The proposed model well predicts the time correlation and the space correlation of the Burgers turbulence.

Existence of Periodic Solutions for Neutral-Type Neural Networks with Delays on Time Scales

In this paper, we employ a fixed point theorem due to Krasnosel’skii to attain the existence of periodic solutions for neutral-type neural networks with delays on a periodic time scale. Some new sufficient conditions are established to show that there exists a unique periodic solution by the contraction mapping principle.

Scale adaptive fitness evaluation‐based particle swarm optimisation for hyperparameter and architecture optimisation in neural networks and deep learning

Research into automatically searching for an optimal neural network(NN)by optimi-sation algorithms is a significant research topic in deep learning and artificial intelligence.However,this is still challenging due to two issues:Both the hyperparameter and ar-chitecture should be optimised and the optimisation process is computationally expen-sive.To tackle these two issues,this paper focusses on solving the hyperparameter and architecture optimization problem for the NN and proposes a novel light‐weight scale‐adaptive fitness evaluation‐based particle swarm optimisation(SAFE‐PSO)approach.Firstly,the SAFE‐PSO algorithm considers the hyperparameters and architectures together in the optimisation problem and therefore can find their optimal combination for the globally best NN.Secondly,the computational cost can be reduced by using multi‐scale accuracy evaluation methods to evaluate candidates.Thirdly,a stagnation‐based switch strategy is proposed to adaptively switch different evaluation methods to better balance the search performance and computational cost.The SAFE‐PSO algorithm is tested on two widely used datasets:The 10‐category(i.e.,CIFAR10)and the 100−cate-gory(i.e.,CIFAR100).The experimental results show that SAFE‐PSO is very effective and efficient,which can not only find a promising NN automatically but also find a better NN than compared algorithms at the same computational cost.

Immunization for scale-free networks by random walker

Based on the random walk and the intentional random walk, we propose two types of immunization strategies which require only local connectivity information. On several typical scale-free networks, we demonstrate that these strategies can lead to the eradication of the epidemic by immunizing a small fraction of the nodes in the networks. Particularly, the immunization strategy based on the intentional random walk is extremely efficient for the assortatively mixed networks.

Cascading failure in the wireless sensor scale-free networks

In the practical wireless sensor networks （WSNs）, the cascading failure caused by a failure node has serious impact on the network performance. In this paper, we deeply research the cascading failure of scale-free topology in WSNs. Firstly, a cascading failure model for scale-free topology in WSNs is studied. Through analyzing the influence of the node load on cascading failure, the critical load triggering large-scale cascading failure is obtained. Then based on the critical load, a control method for cascading failure is presented. In addition, the simulation experiments are performed to validate the effectiveness of the control method. The results show that the control method can effectively prevent cascading failure.

Malicious Code Modeling and Analysis in Weighted Scale-Free Networks

We study the detailed malicious code propagating process in scale-free networks with link weights that denotes traffic between two nodes. It is found that the propagating velocity reaches a peak rapidly then decays in a power-law form, which is different from the well-known result in unweighted network case. Simulation results show that the nodes with larger strength are preferential to be infected, but the hierarchical dynamics are not clearly found. The simulation results also show that larger dispersion of weight of networks leads to slower propagating, which indicates that malicious code propagates more quickly in unweighted scale-free networks than in weighted scale-free networks under the same condition. These results show that not only the topology of networks but also the link weights affect the malicious propagating process.

Long Short-Term Memory Recurrent Neural Network-Based Acoustic Model Using Connectionist Temporal Classification on a Large-Scale Training Corpus

A Long Short-Term Memory(LSTM) Recurrent Neural Network(RNN) has driven tremendous improvements on an acoustic model based on Gaussian Mixture Model(GMM). However, these models based on a hybrid method require a forced aligned Hidden Markov Model(HMM) state sequence obtained from the GMM-based acoustic model. Therefore, it requires a long computation time for training both the GMM-based acoustic model and a deep learning-based acoustic model. In order to solve this problem, an acoustic model using CTC algorithm is proposed. CTC algorithm does not require the GMM-based acoustic model because it does not use the forced aligned HMM state sequence. However, previous works on a LSTM RNN-based acoustic model using CTC used a small-scale training corpus. In this paper, the LSTM RNN-based acoustic model using CTC is trained on a large-scale training corpus and its performance is evaluated. The implemented acoustic model has a performance of 6.18% and 15.01% in terms of Word Error Rate(WER) for clean speech and noisy speech, respectively. This is similar to a performance of the acoustic model based on the hybrid method.

Epidemic spreading in scale-free networks including the effect of individual vigilance

In this paper, we study the epidemic spreading in scale-flee networks and propose a new susceptible-infected- recovered （SIR） model that includes the effect of individual vigilance. In our model, the effective spreading rate is dynamically adjusted with the time evolution at the vigilance period. Using the mean-field theory, an analytical result is derived. It shows that individual vigilance has no effect on the epidemic threshold. The numerical simulations agree well with the analytical result. Purthermore, we investigate the effect of individual vigilance on the epidemic spreading speed. It is shown that individual vigilance can slow the epidemic spreading speed effectively and delay the arrival of peak epidemic infection.

Generalized minimum information path routing strategy on scale-free networks

This paper presents a new routing strategy by introducing a tunable parameter into the minimum information path routing strategy we proposed previously. It is found that network transmission capacity can be considerably enhanced by adjusting the parameter with various allocations of node capability for packet delivery. Moreover, the proposed routing strategy provides a traffic load distribution which can better match the allocation of node capability than that of traditional efficient routing strategies, leading to a network with improved transmission performance. This routing strategy, without deviating from the shortest-path routing strategy in the length of paths too much, produces improved performance indexes such as critical generating rate, average length of paths and average search information.

DIAGNOSTICS OF FATIGUE CRACK IN ULTERIOR PLACES OF LARGER-SCALE OVERLOADED SUPPORTING SHAFT BASED ON TIME SERIES AND NEURAL NETWORKS

To improve the diagnosis accuracy and self-adaptability of fatigue crack in ulterior place of the supporting shaft, time series and neural network are attempted to be applied in research on diag-nosing the fatigue crack’s degree based on analyzing the vibration characteristics of the supporting shaft. By analyzing the characteristic parameter which is easy to be detected from the supporting shaft’s exterior, the time series model parameter which is hypersensitive to the situation of fatigue crack in ulterior place of the supporting shaft is the target input of neural network, and the fatigue crack’s degree value of supporting shaft is the output. The BP network model can be built and net-work can be trained after the structural parameters of network are selected. Furthermore, choosing the other two different group data can test the network. The test result will verify the validity of the BP network model. The result of experiment shows that the method of time series and neural network are effective to diagnose the occurrence and the development of the fatigue crack’s degree in ulterior place of the supporting shaft.