Reparameterization Method Research Articles

Road damage detection algorithm for improved YOLOv5

Road damage detection is an important task to ensure road safety and realize the timely repair of road damage. The previous manual detection methods are low in efficiency and high in cost. To solve this problem, an improved YOLOv5 road damage detection algorithm, MN-YOLOv5, was proposed. We optimized the YOLOv5s model and chose a new backbone feature extraction network MobileNetV3 to replace the basic network of YOLOv5, which greatly reduced the number of parameters and GFLOPs of the model, and reduced the size of the model. At the same time, the coordinate attention lightweight attention module is introduced to help the network locate the target more accurately and improve the target detection accuracy. The KMeans clustering algorithm is used to filter the prior frame to make it more suitable for the dataset and to improve the detection accuracy. To improve the generalization ability of the model, a label smoothing algorithm is introduced. In addition, the structure reparameterization method is used to accelerate model reasoning. The experimental results show that the improved YOLOv5 model proposed in this paper can effectively identify pavement cracks. Compared with the original model, the mAP increased by 2.5%, the F1 score increased by 2.6%, and the model volume was smaller than that of YOLOv5. 1.62 times, the parameter was reduced by 1.66 times, and the GFLOPs were reduced by 1.69 times. This method can provide a reference for the automatic detection method of pavement cracks.

Crown width model for planted Korean pine in eastern Liaoning mountains based on mixed effect linear quantile

Crown width is a critical variable in reflecting the individual tree growth status and in developing forest growth and yield models. With the crown width base model as reference, we developed the crown width quantile regression models for different quantiles (0.50, 0.90, 0.93, 0.95, 0.96, 0.99) based on the data of 2763 Korean pines in 66 permanent plots from the 10-55 years old plantations in Dabiangou forest farm, mountainous areas of eastern Liaoning Province. We used the reparameterization method by introducing the single tree competition index (Rd) and used the dummy variable method by introducing stand density and forest layer variables. We then selected optimal quantile of maximum crown width in the stand by comparing our model developed routine to the traditional methods. The final crown width linear mixed effect quantile regression model was developed based on the optimal quantile at the plot level. The influence of each variable on crown width was analyzed to reflect the difference of crown width among individual trees in the stand. The models with different stand densities and forest layers had significant difference based on F statistical test: the Ra2 of the model increased by 0.0104, the root mean square error decreased by 0.0115 and the mean square error reduction was 7.4%, after the variables of forest layer, forest density, and competition being incorporated into the basic model. The developed quantile regression model performed better than that of the ordinary least square method in simulating the maximum crown width of a single tree in the forest stand. The selected best quantile of the quantile regression model for the upper forest layer and lower forest layer was 0.96 and 0.93, respectively. The linear quantile regression model with the mixed effect was superior to the traditional quantile regression model in Akaike, Bayesion and HQ information criterion and other evaluation para-meters, the standard error for the parameters of estimates was significantly reduced, and the introduced mixed effect well explained differences among different plots. For the upper forest layer and lower forest layer, the maximum crown width decreased with increasing stand density, increased with increasing relative diameters. The influence of stand density on the crown width of the lower forest layer was greater than that of the upper forest layer. The crown width would increase first and then decrease with the increases of DBH when the stand density was large enough. The mixed effect of the quantile regression model developed here could significantly improve the fitting stability of the model. The sustainable development of Korean pine plantation in the mountainous area of eastern Liaoning Pro-vince should be realized by adjusting stand density and moderate tending and thinning in the future.

Robust adaptive neural networks control for a class of time-delay parabolic systems with nonlinear periodic time-varying parameter

This paper mainly focuses on the stabilization of a class of parabolic systems with time-varying delay, external disturbance and nonlinear periodic time-varying parameter (NPTVP). Firstly, combined with Fourier series expansion (FSE) method and neural networks (NNs) approximation technology, a NNs approximator is designed to approximately describe the uncertain dynamic term with NPTVP. Then, based on adaptive control theory, NNs approximation technology and reparameterization method, two robust adaptive neural network control (RANNC) algorithms are designed to make the parabolic systems with time-varying delay, external disturbance and NPTVP achieve asymptotic stability and meet a prescribed adaptive H∞ performance of disturbance attenuation. Sufficient conditions of the asymptotical stability while satisfying the prescribed adaptive H∞ performance index for the resulting closed-loop systems are also derived. Finally, several simulations are carried out to verify the effectiveness of the designed RANNC algorithms.

DANets: Deep Abstract Networks for Tabular Data Classification and Regression

Tabular data are ubiquitous in real world applications. Although many commonly-used neural components (e.g., convolution) and extensible neural networks (e.g., ResNet) have been developed by the machine learning community, few of them were effective for tabular data and few designs were adequately tailored for tabular data structures. In this paper, we propose a novel and flexible neural component for tabular data, called Abstract Layer (AbstLay), which learns to explicitly group correlative input features and generate higher-level features for semantics abstraction. Also, we design a structure re-parameterization method to compress the trained AbstLay, thus reducing the computational complexity by a clear margin in the reference phase. A special basic block is built using AbstLays, and we construct a family of Deep Abstract Networks (DANets) for tabular data classification and regression by stacking such blocks. In DANets, a special shortcut path is introduced to fetch information from raw tabular features, assisting feature interactions across different levels. Comprehensive experiments on seven real-world tabular datasets show that our AbstLay and DANets are effective for tabular data classification and regression, and the computational complexity is superior to competitive methods. Besides, we evaluate the performance gains of DANet as it goes deep, verifying the extendibility of our method. Our code is available at https://github.com/WhatAShot/DANet.

A vector-to-sequence based multilayer recurrent network surrogate model for history matching of large-scale reservoir

History matching can estimate the parameter of spatially varying geological properties and provide reliable numerical models for reservoir development and management. However, in practice, high-dimension, multiple-solutions and computational cost are key issues that restrict the application of history matching methods. Recently, the combination of deep-learning-based surrogate model and sampling algorithm has been widely studied in history matching to overcome the limitations. Considering that real-world large-scale reservoirs often have hundreds of thousands or even millions of grid-based uncertain parameters, extracting spatial features using convolutional neural networks requires a lot of computational cost and storage requirements. Therefore, in this work, we mainly study how to use the recurrent neural network (RNN) to construct the surrogate model for history matching. Specifically, we propose a multilayer RNN surrogate model based on a vector-to-sequence modeling framework. The multilayer RNN surrogate model with gated recurrent unit (GRU), termed MLGRU, is developed to approximate the mapping from feature vector of geological realizations to the production data. The feature vector is the low-dimensional representation of geological parameter fields after using the re-parameterization method, while production data are the simulation results of historical period. In addition, we design a log-transformation-based windowed normalization (LTWN) method for the production data, which can enhance the learnability and features of production data. The MLGRU model is incorporated into a multimodal estimation of distribution algorithm (MEDA) to formulate a history matching workflow. The hyperparameters and performance of the proposed MLGRU model are analyzed by numerical experiments on a 2D reservoir model. Furthermore, numerical experiments performed on the Brugge benchmark model, a large-scale 3D reservoir model, demonstrated the performance of the proposed surrogate model and history matching method.

Limited-Memory Receive Filter Design for Massive MIMO Radar in Signal-Dependent Interference

This letter proposes a limited-memory receive filter design to maximize the output signal-to-interference-plus-noise ratio (SINR) of massive multiple-input multiple-output (MIMO) radar. The associated problem is a minimum variance distortionless response (MVDR) one which can be analytical solved, but the computation of the MVDR solution would be prohibitive in the context of massive MIMO radar, due to the huge memory consumption. To this end, we propose a low-complexity algorithm that exploits a reparameterization method to avoid the significant memory consumption, and solves the resulting problem with several simple algebraic operations. We show that the memory consumption of the proposed algorithm depends on the number of interference sources. Therefore, for the scenario with a large number of interference sources, we extend the proposed algorithm by the alternating direction method of multipliers (ADMM), which promotes a further memory reduction. Finally, numerical simulations are conducted to demonstrate the efficiency of the proposed algorithm.

Distributed consensus control of periodically time-varying multi-agent systems using neural networks and fourier series expansion

This paper studies the consensus of nonlinear multi-agent systems with periodic disturbances and uncertain dynamics based on matrix theory, adaptive control, neural networks and fourier series expansion. Firstly, fourier series expansion and neural networks are used to describe the unknown periodic time-varying parameter and uncertain nonlinear dynamic, respectively. Secondly, based on adaptive control technology and reparameterization method, two new fully distributed control protocols are designed based on symbolic function and smooth hyperbolic tangent function, respectively, so that all agents can reach asymptotic consensus. Thirdly, a new positive integral bounded function is introduced to compensate for the approximation error caused by the smooth hyperbolic tangent function instead of the symbolic function, so that all network nodes achieve the same consensus effect. Finally, a simulation example is given to verify the effectiveness of the two algorithms and to illustrate their advantages and disadvantages.

Revised Atomic Charges for OPLS Force Field Model of Poly(Ethylene Oxide): Benchmarks and Applications in Polymer Electrolyte.

Poly(ethylene oxide) (PEO)-based polymers are common hosts in solid polymer electrolytes (SPEs) for high-power energy devices. Molecular simulations have provided valuable molecular insights into structures and ion transport mechanisms of PEO-based SPEs. The calculation of thermodynamic and kinetic properties rely crucially on the dependability of the molecular force fields describing inter- and intra-molecular interactions with the target system. In this work, we reparametrized atomic partial charges for the widely applied optimized potentials for liquid simulations (OPLS) force field of PEO. The revised OPLS force field, OPLSR, improves the calculations of density, thermal expansion coefficient, and the phase transition of the PEO system. In particular, OPLSR greatly enhances the accuracy of the calculated dielectric constant of PEO, which is critical for simulating polymer electrolytes. The reparameterization method was further applied to SPE system of PEO/LiTFSI with O:Li ratio of 16:1. Based on the reparametrized partial charges, we applied separate charge-scaling factors for PEO and Li salts. The charge-rescaled OPLSR model significantly improves the resulting kinetics of Li+ transport while maintaining the accurate description of coordination structures within PEO-based SPE. The proposed OPLSR force field can benefit the future simulation studies of SPE systems.

JDGAN: Enhancing generator on extremely limited data via joint distribution

Generative Adversarial Network (GAN) is a thriving generative model and considerable efforts have been made to enhance the generation capabilities via designing a different adversarial framework of GAN (e.g., the discriminator and the generator) or redesigning the penalty function. Although existing models have been demonstrated to be very effective, their generation capabilities have limitations. Existing GAN variants either result in identical generated instances or generate simulation data with low quality when the training data are diverse and extremely limited (a dataset consists of a set of classes but each class holds several or even one single sample) or extremely imbalanced (a category holds a set of samples and other categories hold one single sample). In this paper, we present an innovative approach to tackle this issue, which jointly employs joint distribution and reparameterization method to reparameterize the randomized space as a mixture model and learn the parameters of this mixture model along with that of GAN. In this way, we term our approach Joint Distribution GAN (JDGAN). In our work, we show that the JDGAN can not only generate high quality simulation data with diversity, but also increase the overlapping area between the generating distribution and the raw data distribution. We proceed to conduct extensive experiments, utilizing MNIST, CIFAR10 and Mass Spectrometry datasets, all using extremely limited amounts of data, to demonstrate the significant performance of JDGAN in both achieving the smallest Fréchet Inception Distance (FID) score and producing diverse generated data.

Forecasting fish yield using statistical nonlinear growth models – A reparameterization concept

The concept of reparameterization is rarely employed in forecasting fish yield, even though reparameterization method is computationally more efficient especially when estimation involves an iterative method. Further, a reparameterization of the growth functions whose new parameters correspond to scientifically interesting characteristics of the growth process (Choi et al, 2009). Reparameterization of the same basic model produce the same goodness-of-fit and the same fitted values, etc. but they may differ greatly in their estimation behavior. Further, a class of parameters that exhibits close-to-linear behavior is the class of expected-value parameters. The extent of non-normality and excess variance of the least squares estimates of an expected value parameter is generally small (Ratkowsky, 1990). The only restriction on expected-value parameters is that they should fall within the observed range of the data and not correspond to asymptotes or extrapolations outside the data range. Expected-value parameters outside the range of the observed data are less efficacious. Also, the derived expressions due to expected-value parameters are usually more cumbersome in appearance than the original expressions, and the expected-value parameters may appear more than once. As compensation for this loss of aesthetics, however, parameterizations with expected-value parameters offer three advantages. First, one obtains rapid convergence to the leastsquares estimates using the new parameterization, since the new model close-to-linear. A second benefit is that initial parameter estimates are very easy to obtain. Third, the expected-value parameters are more suitable for inference than the original parameters since their least-squares estimates are close-to being unbiased, normally distributed, minimum variance estimators. Although the role of expected-value parameters is very important, algebraic limitations prevent the universal application of expected-value parameters. However, in most of the cases a few parameters (one or two parameters only) are responsible for the far-from-linear behavior of a nonlinear regression model, it is seldom necessary to replace all the parameters of a model by expected-value parameters (Ratkowsky, 1990). Singh (2011) derived an explicit expression for partial reparameterization of Schaefer model with autoregressive of order one using expected-value parameters and fitted to catch-effort fishery data.

A highly accurate boundary integral equation method for surfactant-laden drops in 3D

The presence of surfactants alters the dynamics of viscous drops immersed in an ambient viscous fluid. This is specifically true at small scales, such as in applications of droplet based microfluidics, where the interface dynamics become of increased importance. At such small scales, viscous forces dominate and inertial effects are often negligible. Considering Stokes flow, a numerical method based on a boundary integral formulation is presented for simulating 3D drops covered by an insoluble surfactant. The method is able to simulate drops with different viscosities and close interactions, automatically controlling the time step size and maintaining high accuracy also when substantial drop deformation appears. To achieve this, the drop surfaces as well as the surfactant concentration on each surface are represented by spherical harmonics expansions. A novel reparameterization method is introduced to ensure a high-quality representation of the drops also under deformation, specialized quadrature methods for singular and nearly singular integrals that appear in the formulation are evoked and the adaptive time stepping scheme for the coupled drop and surfactant evolution is designed with a preconditioned implicit treatment of the surfactant diffusion.

Kinematically smoothing trajectories by NURBS reparameterization – an innovative approach

This paper presents an innovative curve reparameterization method for kinematically smoothing trajectories based on jerk minimization. In this research, not only the trajectory is expressed in terms of a parametric NURBS curve, but also the curve parameter is related to the execution time by means of a NURBS reparameterization function. Meanwhile, an analytical sensitivity analysis is developed to minimize the jerk magnitude of the trajectory in a gradient-based optimization. The advantage of the proposed methodology is that the geometric shape of the trajectory’s path remains unchanged while the jerk is smoothed. Hence, the trajectory will be tracked by the optimized values of velocity, acceleration, and jerk. It would be helpful when the trajectory’s path cannot be changed because of the geometrical limitations or obstacle avoidance scheme. The applicability of the presented approach is also illustrated by some case studies.

Using hierarchical centering to facilitate a reversible jump MCMC algorithm for random effects models

Hierarchical centering has been described as a reparameterization method applicable to random effects models. It has been shown to improve mixing of models in the context of Markov chain Monte Carlo (MCMC) methods. A hierarchical centering approach is proposed for reversible jump MCMC (RJMCMC) chains which builds upon the hierarchical centering methods for MCMC chains and uses them to reparameterize models in an RJMCMC algorithm. Although these methods may be applicable to models with other error distributions, the case is described for a log-linear Poisson model where the expected value λ includes fixed effect covariates and a random effect for which normality is assumed with a zero-mean and unknown standard deviation. For the proposed RJMCMC algorithm including hierarchical centering, the models are reparameterized by modeling the mean of the random effect coefficients as a function of the intercept of the λ model and one or more of the available fixed effect covariates depending on the model. The method is appropriate when fixed-effect covariates are constant within random effect groups. This has an effect on the dynamics of the RJMCMC algorithm and improves model mixing. The methods are applied to a case study of point transects of indigo buntings where, without hierarchical centering, the RJMCMC algorithm had poor mixing and the estimated posterior distribution depended on the starting model. With hierarchical centering on the other hand, the chain moved freely over model and parameter space. These results are confirmed with a simulation study. Hence, the proposed methods should be considered as a regular strategy for implementing models with random effects in RJMCMC algorithms; they facilitate convergence of these algorithms and help avoid false inference on model parameters.

A new method to improve estimation of uncertain parameters in the Ensemble Kalman filter by re-parameterization employing prior statistics correction

The Ensemble Kalman Filter (EnKF) is a Monte Carlo based method to assimilate the measurement data sequentially in time. Although, EnKF has some advantages over the other Kalman based methods to deal with non-linear and/or high dimensional reservoir models, it also suffers from deficiency in estimation of non-Gaussian parameters. In this work, we propose a re-parameterization method to handle non-Gaussian parameters via Ensemble Kalman Filter framework. For this purpose, concept of cumulative distribution function transformation has been used. In addition, the statistics of prior information have been aggregated in the state vector in order to capture the prior uncertainties of non-Gaussian parameters. To evaluate the performance of the new method, three estimation examples have been implemented. These examples are designed to evaluate the performance of the proposed method to estimate both local and global parameters. The performance of method in handling non-Gaussian prior as well as capturing the prior uncertainty has been assessed. The results revealed that, the proposed method handled both non-Gaussian prior and capturing of the prior uncertainties of the local and global parameters quite efficiently. Compared to previously conducted researches, new algorithm would considerably reduce the number of required ensemble members to converge to the appropriate solution. Standard ensemble Kalman filter and normal score transformation have been also implemented for each setup; the obtained results show that the new proposed method outperforms the other alternate methods.

Meaningful aspects of change as novel random coefficients: a general method for reparameterizing longitudinal models.

A fundamental goal of longitudinal modeling is to obtain estimates of model parameters that reflect meaningful aspects of change over time. Often, a linear or nonlinear model may be sensible from a theoretical perspective, yet may have parameters that are difficult to interpret in a way that sheds light on substantive hypotheses. Fortunately, such models may be reparameterized to yield more easily interpretable parameters. This article has 3 goals. First, we provide theoretical background and elaboration on Preacher and Hancock's (2012) 4-step method for reparameterizing growth curve models. Second, we extend this method by providing a user-friendly modification of the structured latent curve model in the third step that enables fitting models that are not estimable with the original method. This modification also allows researchers to specify the mean structure without having to determine which parameters enter nonlinearly and without needing to solve complex matrix expressions. Third, we illustrate how this general reparameterization method allows researchers to treat the average rate of change, half-life, and knot (transition point) as random coefficients; these aspects of change have not before been treated as random coefficients in structural equation modeling. We supply Mplus code for illustrative examples in an online supplement. Our core message is that growth curve models are considerably more flexible than most researchers may suspect. Virtually any parameter can be treated as a random coefficient that varies across individuals. Alternative parameterizations of a given model may yield unique insights that are not available with traditional parameterizations.

High-quality construction of analysis-suitable trivariate NURBS solids by reparameterization methods

High-quality volumetric parameterization of computational domain plays an important role in three-dimensional isogeometric analysis. Reparameterization technique can improve the distribution of isoparametric curves/surfaces without changing the geometry. In this paper, using the reparameterization method, we investigate the high-quality construction of analysis-suitable NURBS volumetric parameterization. Firstly, we introduce the concept of volumetric reparameterization, and propose an optimal Mobius transformation to improve the quality of the isoparametric structure based on a new uniformity metric. Secondly, from given boundary NURBS surfaces, we present a two-stage scheme to construct the analysis-suitable volumetric parameterization: in the first step, uniformity-improved reparameterization is performed on the boundary surfaces to achieve high-quality isoparametric structure without changing the shape; in the second step, from a new variational harmonic metric and the reparameterized boundary surfaces, we construct the optimal inner control points and weights to achieve an analysis-suitable NURBS solid. Several examples with complicated geometry are presented to illustrate the effectiveness of proposed methods.

Mapped B-spline basis functions for shape design and isogeometric analysis over an arbitrary parameterization

It is well-known that B-spline surfaces are defined by a regular array of control vertices. In case of models with arbitrary topology, it is extremely difficult to maintain continuity conditions among neighboring surfaces. The scenario is also true for applications in isogeometric analysis (IGA). This paper presents a novel method for shape design and isogeometric analysis from a quadrilateral control mesh of arbitrary topology using mapped B-spline basis functions. Based on an arbitrary input quadrilateral control mesh, a global parameterization of the final surface is first defined through a Gravity Center Method (GCM). A re-parameterization method is then applied to map a B-spline basis function to others that are explicitly defined and are tailored to each of the control vertices which can be either regular or extraordinary ones. The final surface is defined by all the input control vertices with their corresponding mapped basis functions. For practical implementation, the surface can be evaluated patch by patch. Depending on the order of the B-spline basis function used for mapping to others, the global continuity of the resulting surface, including at extraordinary points, can be arbitrary higher order. In the present paper, a uniform cubic B-spline basis function is used and the resulting surface is globally C2 continuous. Several numerical examples are provided to demonstrate the proposed method for both shape design and isogeometric analysis.

A New Piecewise Rational Reparameterization Method

A new piecewise rational re-parameterization of a polynomial curve is presented. The method relies on a parametric polygonal decomposition method. With the proposed piecewise rational re-parameterization method, the optimal parameterization can be reached, and the optimal parameterizations asymptotic convergence to the arc-length parameterization is analyzed. Computing instances demonstrates the efficiency of the proposed method.

Non-certainty Equivalent Adaptive Exciting Control of Multi-machine Power Systems

Transient stability problem for multi-machine infinite bus system with the generator excitation was addressed via the non-certainty equivalent nonlinear re-parameterization method. The system need not to be linearized. The damping coefficient uncertainty was considered. A non-certainty equivalent excitation controller and a novel parameter updating law were obtained simultaneously via adaptive backstepping and Lyapunov methods to achieve stability of the error systems. Simulation results showed that the proposed controller had good transient performance. DOI: http://dx.doi.org/10.11591/telkomnika.v11i7.285 9

An RBF-Based Reparameterization Method for Constrained Texture Mapping

Texture mapping has long been used in computer graphics to enhance the realism of virtual scenes. However, to match the 3D model feature points with the corresponding pixels in a texture image, surface parameterization must satisfy specific positional constraints. However, despite numerous research efforts, the construction of a mathematically robust, foldover-free parameterization that is subject to positional constraints continues to be a challenge. In the present paper, this foldover problem is addressed by developing radial basis function (RBF)-based reparameterization. Given initial 2D embedding of a 3D surface, the proposed method can reparameterize 2D embedding into a foldover-free 2D mesh, satisfying a set of user-specified constraint points. In addition, this approach is mesh free. Therefore, generating smooth texture mapping results is possible without extra smoothing optimization.