Network Decomposition Research Articles

Self-supervised monocular depth estimation based on improved dense network and wavelet decomposition

Traditional self-supervised monocular depth estimation models do not adequately extract and fuse shallow features, which can easily lead to problems such as missed detection of small objects and blurred object edges. To address the above problems, this paper proposes a self-supervised monocular depth estimation model based on improved dense network and wavelet decomposition. The model follows the structure of U-net, in which the encoder adopts an improved dense connection method to improve the feature extraction and fusion capabilities of the encoder; at the same time, a detail enhancement module is added to the jump connection to further refine and integrate the multi-scale features output by the encoder; finally, the wavelet decomposition method is introduced in the decoder, forcing the decoder to pay more attention to high-frequency information and achieve refined processing of image edges. Experimental results show that the depth estimation model proposed in this paper has a stronger ability to capture small object features, and the edges of the generated depth map are clearer and more accurate.

Predicting e-commerce product prices through the integration of variational mode decomposition and deep neural networks

Product prices frequently manifest nonlinear and nonstationary time-series attributes, indicating potential variations in their behavioral patterns over time. Conventional linear models may fall short in adequately capturing these intricate properties. In addressing this, the present study leverages the adaptive and non-recursive attributes of the Variational Mode Decomposition (VMD) methodology. It employs VMD to dissect the intricate time series into multiple Intrinsic Mode Functions (IMF). Subsequently, a method rooted in the minimum fuzzy entropy criterion is introduced for determining the optimal modal number (K) in the VMD decomposition process. This method effectively mitigates issues related to modal confusion and endpoint effects, thereby enhancing the decomposition efficacy of VMD. In the subsequent phase, deep neural networks (DNN) are harnessed to forecast the identified modes, with the cumulative modal predictions yielding the ultimate e-commerce product price prognostications. The predictive efficacy of the proposed Variational Mode Decomposition-deep neural network (VMD-DNN) decomposition model is assessed on three public datasets, wherein the mean absolute percentage error (MAPE) on the E-commerce Price Prediction Dataset and Online Retail Dataset is notably low at 0.6578 and 0.5414, respectively. This corresponds to a remarkable error reduction rate of 66.5% and 70.4%. Moreover, the VMD-DNN decomposition model excels in predicting e-commerce product prices through DNN, thereby amplifying the VMD decomposition capability by 4%. The VMD-DNN model attains superior results in terms of directional symmetry, boasting the highest Directional Symmetry (DS) score of 86.25. Notably, the forecasted trends across diverse price ranges closely mirror the actual trends.

Optimizing Fleet Size in Point-to-Point Shared Demand Responsive Transportation Service: A Network Decomposition Approach

With increasing urbanization and the demand for efficient, flexible transportation solutions, demand-responsive transportation services (DTRS) has emerged as a viable alternative to traditional public transit. However, determining the optimal fleet size to balance the investment and operational revenue remains a significant challenge for service providers. In this article, we address the optimization of fleet size in point-to-point shared demand DRTS, which widely operates within many cities. To capture the uncertain passenger demands in the future when planning the fleet size currently, we model this problem with a framework of two-stage stochastic programming with recourse. Fleet sizing decisions are made in the first stage before the uncertain demands are revealed. After the uncertainty is revealed, the second stage involves making additional decisions to maximize operational revenue. The objective is to optimize the total revenue of the first-stage decisions and the expected revenue of the recourse actions. To solve this practical problem, we resort to the Model Predictive Control method (MPC) and propose a network decomposition approach that first converts the transportation network to a nodal tree structure and then develops a Nodal Tree Recourse with Dependent Arc Capacities (NTRDAC) algorithm to obtain the exact value of the expected recourse functions. In the experiments, NTRDAC is able to produce results within seconds for transportation networks with over 30 nodes. In contrast, a commercial solver is only capable of solving networks with up to five nodes. The stability tests show that NTRDAC remains robust as the problem size varies. Lastly, the value of the stochastic solution (VSS) was evaluated, and the results indicate that it consistently outperforms the expected value solutions. Numerical experiments show that the performance of the NTRDAC algorithm is quite encouraging and fit for large-scale practical problems.

Autonomous radar interference detection and mitigation using neural network and signal decomposition

<p>Autonomous radar interference is a challenging problem in autonomous vehicle systems. Interference signals can decrease the signal-to-interference-noise ratio (SINR), and this condition decreases the performance detection of autonomous radar. This paper exploits a neural network and signal decomposition to detect and mitigate radar interference in autonomous vehicle applications. A neural network (NN) with four inputs, one hidden layer, and one output is trained with various signal-to-noise (SNR), interference radar bandwidth, and sweep time of autonomous radar. Four inputs of NN represent SNR, mean, total harmonic distortion (THD), and root means square (RMS) of the received radar signal. Variational mode decomposition (VMD) and zeroing based on a constant false alarm rate (CFAR-Z) are used to mitigate radar interference. VMD algorithm is applied to decompose interference signals into multi-frequency sub-band. As a result, the proposed neural network can detect radar interference, and NN-VMD-CFAR-Z can increase SINR up to 2dB higher than the NN-CFAR-Z algorithm.</p>

Modelling network motifs as higher order interactions: a statistical inference based approach

The prevalent approach to motif analysis seeks to describe the local connectivity structure of networks by identifying subgraph patterns that appear significantly more often in a network then expected under a null model that conserves certain features of the original network. In this article we advocate for an alternative approach based on statistical inference of generative models where nodes are connected not only by edges but also copies of higher order subgraphs. These models naturally lead to the consideration of latent states that correspond to decompositions of networks into higher order interactions in the form of subgraphs that can have the topology of any simply connected motif. Being based on principles of parsimony the method can infer concise sets of motifs from within thousands of candidates allowing for consistent detection of larger motifs. The inferential approach yields not only a set of statistically significant higher order motifs but also an explicit decomposition of the network into these motifs, which opens new possibilities for the systematic study of the topological and dynamical implications of higher order connectivity structures in networks. After briefly reviewing core concepts and methods, we provide example applications to empirical data sets and discuss how the inferential approach addresses current problems in motif analysis and explore how concepts and methods common to motif analysis translate to the inferential framework.

Deformation Mechanisms Dominated by Decomposition of an Interfacial Misfit Dislocation Network in Ni/Ni3Al Multilayer Structures.

Ni/Ni3Al heterogeneous multilayer structures are widely used in aerospace manufacturing because of their unique coherent interfaces and excellent mechanical properties. Revealing the deformation mechanisms of interfacial structures is of great significance for microstructural design and their engineering applications. Thus, this work aims to establish the connection between the evolution of an interfacial misfit dislocation (IMD) network and tensile deformation mechanisms of Ni/Ni3Al multilayer structures. It is shown that the decomposition of IMD networks dominates the deformation of Ni/Ni3Al multilayer structures, which exhibits distinct effects on crystallographic orientation and layer thickness. Specifically, the Ni/Ni3Al (100) multilayer structure achieves its maximum yield strength of 5.28 GPa at the layer thickness of 3.19 nm. As a comparison, the (110) case has a maximum yield strength of 4.35 GPa as the layer thickness is 3.01 nm. However, the yield strength of the (111) one seems irrelevant to layer thickness, which fluctuates between 10.89 and 11.81 GPa. These findings can provide new insights into a deep understanding of the evolution and deformation of the IMD network of Ni/Ni3Al multilayer structures.

Bayesian learning of gas transport in three-dimensional fracture networks

Modeling gas flow through fractures of subsurface rock is a particularly challenging problem because of the heterogeneous nature of the material. High-fidelity simulations using discrete fracture network (DFN) models are one methodology for predicting gas particle breakthrough times at the surface but are computationally demanding. We propose a Bayesian machine learning method that serves as an efficient surrogate model, or emulator, for these three-dimensional DFN simulations. Our model trains on a small quantity of simulation data with given statistical properties and, using a graph/path-based decomposition of the fracture network, rapidly predicts quantiles of the breakthrough time distribution on DFNs with those statistical properties. The approach, based on Gaussian Process Regression (GPR), outputs predictions that are within 20%–30% of high-fidelity DFN simulation results. Unlike previously proposed methods, it also provides uncertainty quantification, outputting confidence intervals that are essential given the uncertainty inherent in subsurface modeling. Our trained model runs within a fraction of a second, considerably faster than reduced-order models yielding comparable accuracy (Hyman et al., 2017; Karra et al., 2018) and multiple orders of magnitude faster than high-fidelity simulations.

Spectral Triadic Decompositions of Real-World Networks

Spectral Triadic Decompositions of Real-World Networks

Resource Optimization for Quantum Dynamics with Tensor Networks: Quantum and Classical Algorithms.

The exponential scaling of the quantum degrees of freedom with the size of the system is one of the biggest challenges in computational chemistry and particularly in quantum dynamics. We present a tensor network approach for the time-evolution of the nuclear degrees of freedom of multiconfigurational chemical systems at a reduced storage and computational complexity. We also present quantum algorithms for the resultant dynamics. To preserve the compression advantage achieved via tensor network decompositions, we present an adaptive algorithm for the regularization of nonphysical bond dimensions, preventing the potentially exponential growth of these with time. While applicable to any quantum dynamical problem, our method is particularly valuable for dynamical simulations of nuclear chemical systems. Our algorithm is demonstrated using ab initio potentials obtained for a symmetric hydrogen-bonded system, namely, the protonated 2,2'-bipyridine, and compared to exact diagonalization numerical results.

Multi-Dimensional Data Recovery via Feature-Based Fully-Connected Tensor Network Decomposition

Multi-Dimensional Data Recovery via Feature-Based Fully-Connected Tensor Network Decomposition

Nonparametric inference of higher order interaction patterns in networks

Local interaction patterns play an important role in the structural and functional organization of complex networks. Here we propose a method for obtaining parsimonious decompositions of networks into higher order interactions which can take the form of arbitrary motifs. The method is based on a class of analytically solvable generative models which in combination with non-parametric priors allow us to infer higher order interactions from dyadic graph data without any prior knowledge on the types or frequencies of such interactions. We test the presented approach on simulated data for which we recover the set of underlying higher order interactions to a high degree of accuracy. For empirical networks the method identifies concise sets of atomic subgraphs from within thousands of candidates that cover a large fraction of edges and include higher order interactions of known structural and functional significance. Being based on statistical inference the method also produces a fit of the network to analytically tractable higher order models opening new avenues for the systematic study of higher order interactions.

Structural analysis and the sum of nodes’ betweenness centrality in complex networks

Structural analysis in the field of network science aims to uncover the hidden information embedded within the topological structure of complex networks. Betweenness centrality is a measure that quantifies a node’s influence on the overall network, based on the shortest paths between different pairs of nodes. This unique property of betweenness centrality allows it to capture more structural information than the commonly employed degree centrality, which is a fundamental characteristic of network structure. In this study, we demonstrate that the sum of nodes’ betweenness centralities (SBC) can be utilized as a novel structural index to reveal the underlying rules governing the growth of a network. Additionally, we have developed a method that combines K-shell decomposition and SBC analysis to investigate the growth rules that have shaped the evolution of static networks in the past. Our findings indicate that the Barabási–Albert model guides the network’s SBC to grow in a logarithmic fashion, whereas the Erdős–Rényi model leads to the convergence of SBC values. Interestingly, we also discover that the convergence or divergence of SBC within the k-shell-like decomposition of a static network can be used to distinguish between developed networks (where growth rules have reached saturation) and developing networks (still undergoing expansion). These results highlight the utility of SBC as a reasonable and effective index for the structural analysis of complex networks, providing insights into the rules governing their evolution.

SVD-based algorithms for fully-connected tensor network decomposition

SVD-based algorithms for fully-connected tensor network decomposition

Non-Local and Fully Connected Tensor Network Decomposition for Remote Sensing Image Denoising

Non-Local and Fully Connected Tensor Network Decomposition for Remote Sensing Image Denoising

Optimal multimodal multi-echelon vaccine distribution network design for low and medium-income countries with manufacturing infrastructure during healthcare emergencies

In the event of routine immunizations for containing disease outbreaks, the Expanded Programme on Immunization (EPI) led Low and Medium-Income Countries (LMICs) follow a four-tier distribution network. Such a network is suitable for LMICs importing vaccines rather than manufacturing and may not be suitable for healthcare emergencies. Recent shift from routine immunizations to pandemic-orientation, calls for large-scale multimodal distribution models. This paper proposes a cold chain network for LMICs with manufacturing hubs during health emergencies, integrating vaccine manufacturers and incorporating multiple transportation modes. The problem is formulated as a Mixed Integer Quadratically Constrained Programming (MIQCP) problem, to minimize ‘Cumulative Transportation Time’ (CTT) with the minimal number of trips. To solve large-scale problems the paper also proposes a novel algorithm that divides certain parts of the network and retains the edges that are likely to give better solutions. The proposed model extends the existing body of the literature by introducing features such as vehicle choice between locations, generalized hierarchical structure, network decomposition, and appropriateness with LMIC networks. We conduct numerical experiments to validate the proposed algorithm using data from India. The results show that the proposed model saves cumulative transportation time along the optimally generated network when compared with the actual vaccine supply network in India during healthcare emergencies. The proposed algorithm is also faster than the formulation giving comparable results while solved in a commercial solver.

Novel sol-gel route for expanding the glass forming region of tin phosphate glass for secondary battery electrode applications

In this paper, tin-phosphate amorphous glass and nano-crystallized ceramic materials were synthesized by sol-gel methods and heat treated at 400 °C with the composition of xSnO2− (100-x)P2O5 (xSnP: ‘x’ from 40 to 90 mol%). The electrical and thermal properties of tin-phosphate materials were investigated by electrical impedance spectroscopy (EIS) and thermogravimetry – differential thermal analysis – mass spectroscopy (TG–DTA–MS). Moreover, the local structure of xSnP was characterized by 119Sn-Mössbauer and magic angle spinning (MAS) 31P NMR spectroscopies. Homogeneous glasses were obtained for xSnP with ‘x’ from 50 to 80, while Sn(HPO4), SnO2, and Sn3(PO4)2 nanoparticles formed in 40SnP and 90SnP compositions, respectively. The nano-crystallized 40SnP and 90SnP samples showed large electrical conductivity (σ) of 2.02×10−5 and 1.04×10−4S cm−1 due to the conductive crystalline phases of SnP2O7 and defective SnO2, respectively. The lower σ of about 10−8S cm−1 were recorded for xSnP with ‘x’ from 50 to 80 owing to the decomposition of the 3D network built by phosphate tetrahedrons. 119Sn-Mössbauer spectra proved the growing number of SnIVO6 units, which do not contribute to the rise of electrical conductivity. On the other hand, larger σ values of 1.79×10−4 and 6.4×10−3S cm−1 were confirmed for 40SnP and 90SnP samples after heat treatment, resulting in partial reduction of SnIV to SnII and growing of the 3D network structure by PO43−. It is concluded that highly electrically conductive xSnP samples can be synthesized by the sol-gel method followed by slow heat treatment. In addition, the battery evaluation was performed of xSnP samples as cathode and anode in a sodium-ion battery. The highest initial capacity recorded for 90SnP nano-crystallized ceramic at 37.97 mAh g−1 under a current rate of 20 mA g−1 with capacity retention (%) of 48.15% after 100 cycles as a cathode material. The anode test showed that the 40SnP nano-crystallized ceramic has the largest initial capacity of 375 mAh g−1 and after heat treatment 50SnP glass significantly increased to 425 mAh g−1 compared with 186 mAh g−1 before heat treatment. Also, the heat treatment effect on the initial capacity of 90SnP nano-crystallized ceramic with observation largest value of 470 mAh g−1. These results open the way to maybe applying these glass and nano-crystallized ceramic materials in all-solid-state batteries.

Study of using cork powder as an adjuvant bio-flame retardant in acrylonitrile-butadiene-styrene flame retardant formulations

By having a challenging prospect of developing more environmentally friendly flame-retardant (FR) acrylonitrile-butadiene-styrene (ABS) formulations, an initial study related to the effect of combining ammonium polyphosphate (APP) with cork powder (C), a bio-waste of the cork industry on ABS-FR thermal stability and fire performance is proposed and discussed in the present work. ABS composites with a global filler weight content of 30 wt.% were prepared by using a melt-compounding process. Two different types of thermal decomposition behavior, assessed by means of thermogravimetric analysis (TGA), were observed for ABS/C/APP composites. For formulations with C/APP weight ratio < 1, the thermal decomposition of APP polyphosphate network was shifted to lower temperature. On the other hand, when C/APP ratio was ≥ 1, a new thermal decomposition (TD) step related to an earlier thermal decomposition of cork components was registered between 320 °C and 370 °C. Moreover, by means of Fourier-transform infrared spectroscopy (FTIR) analysis of the pyrolysis gasses, a broader temperature range of ammonia and water release was noticed when APP and C were simultaneously incorporated in ABS. In addition, a micro-scale combustion calorimeter (MCC) analysis revealed that replacing APP by cork led to a decrease of the heat release capacity (HRC) and an increase of the residue formation compared to when only APP was present in ABS composite. Furthermore, forced flaming fire characterization of ABS formulations, studied by means of cone calorimetry (CC), revealed that replacing 20 wt.% of APP by cork (ABS/20C/10APP) led to a similar fire performance to the ABS formulation with 30 wt.% of APP (ABS/30APP). A similar trend under horizontal UL-94 was observed, nevertheless, for the composite with cork in its composition no dripping was noticed. Additionally, by only replacing a 3 wt.% of APP (ABS/3C/27APP) by cork an optimum improvement in fire retardancy was achieved, with the highest efficiency in reducing the peak heat release rate (pHRR) of 21 % regarding ABS/30APP. These results have been related to the chemical interaction between APP and cork components in the condensed phase confirmed by the observance of a more intense P-O-C absorbance peak of the CC residue of ABS/3C/27APP, analyzed by means of FTIR. This composite also showed a more cohesive and expanded char layer, observed by scanning electron microscopy (SEM), which acted as a more efficient protective barrier, reducing heat and mass transfer phenomena during combustion which also led to the highest reduction in linear burning rate (LBR) under horizontal UL-94.

Global optimization strategy of prosumer data center system operation based on multi-agent deep reinforcement learning

The escalating issues of high energy consumption and carbon emissions in data centers (DCs) necessitate the optimization of system operations. However, early optimization strategies were overly simplistic and lacked automated updating and iterative capabilities. With the evolution of artificial intelligence (AI), researchers have applied deep reinforcement learning (DRL) algorithms to system operations. However, the optimization focus has been limited to the internal systems, lacking global optimization. In this paper, a global optimization control strategy based on the Dueling double-deep Q network (D3QN) and value decomposition network (VDN) algorithms is proposed to make the DCs system operate more closely with the upstream, midstream, and downstream. By adjusting battery charging/discharging capacity, computational workload, and waste heat utilization heating temperature global synergistic optimization is achieved. Compared with without optimization, renewable energy waste, operation cost, total electricity consumption, and grid electricity consumption are reduced by 18.37%, 9.78%, 4.01%, and 29.74%, respectively. Additionally, a detailed comparison between non-algorithmic optimization and algorithmic optimization is provided, offering valuable insights for substantial energy savings and emissions reduction in DCs. The results demonstrate the importance of fully exploring the interactive potential between upstream energy supply, midstream computational workload, and downstream waste heat recovery to achieve synergistic global optimization of “computing power", “thermal energy" and “electrical energy" for the sustainable and green development of DCs or other prosumer buildings.

Tensor network decompositions for absolutely maximally entangled states

Absolutely maximally entangled (AME) states of k qudits (also known as perfect tensors) are quantum states that have maximal entanglement for all possible bipartitions of the sites/parties. We consider the problem of whether such states can be decomposed into a tensor network with a small number of tensors, such that all physical and all auxiliary spaces have the same dimension D. We find that certain AME states with k=6 can be decomposed into a network with only three 4-leg tensors; we provide concrete solutions for local dimension D=5 and higher. Our result implies that certain AME states with six parties can be created with only three two-site unitaries from a product state of three Bell pairs, or equivalently, with six two-site unitaries acting on a product state on six qudits. We also consider the problem for k=8, where we find similar tensor network decompositions with six 4-leg tensors.

A random sampling algorithm for fully-connected tensor network decomposition with applications

A random sampling algorithm for fully-connected tensor network decomposition with applications