In order to solve the problem that the ORB algorithm increases the probability of feature point loss and mis-matching in some cases such as insufficient light intensity, low texture, large camera rotation, etc. This p...
详细信息
ISBN:
(纸本)9783031138706;9783031138690
In order to solve the problem that the ORB algorithm increases the probability of feature point loss and mis-matching in some cases such as insufficient light intensity, low texture, large camera rotation, etc. This paper introduces an enhanced graphical local adaptive thresholding (EGLAT) feature extraction algorithm, which enhances the front-end real-time input image to make the blurred texture and corners clearer, replacing the existing ORB extraction method based on static thresholding, the local adaptive thresholding algorithm makes the extraction of feature points more uniform and good quality, avoiding the problems of over-concentration of feature points and partial information loss. Comparing the proposed algorithm with ORB-SLAM2 in a public dataset and a real environment, the results show that our proposed method outperforms the ORB-SLAM2 algorithm in terms of the number of extracted feature points, the correct matching rate and the matching time, especially the matching rate of feature points is improved by 18.7% and the trajectory error of the camera is reduced by 16.5%.
Nonconvex optimization problems have always been one focus in deep learning, in which many fast adaptive algorithms based on momentum are applied. However, the full gradient computation of high-dimensional feature vec...
详细信息
Nonconvex optimization problems have always been one focus in deep learning, in which many fast adaptive algorithms based on momentum are applied. However, the full gradient computation of high-dimensional feature vector in the above tasks become prohibitive. To reduce the computation cost for optimizers on nonconvex optimization problems typically seen in deep learning, this work proposes a randomized block -coordinate adaptive optimization algorithm, named RAda, which randomly picks a block from the full coordinates of the parameter vector and then sparsely computes its gradient. We prove that RAda converges to a 6-accurate solution with the stochastic first-order complexity of O(1/62), where 6 is the upper bound of the gradient's square, under nonconvex cases. Experiments on public datasets including CIFAR-10, CIFAR-100, and Penn TreeBank, verify that RAda outperforms the other compared algorithms in terms of the computational cost.
作者:
Xing, ZhaoyuTan, HuaiyuZhong, WeiShi, LeiXiamen Univ
Paula & Gregory Chow Inst Studies Econ 422 Siming South Rd Xiamen 361005 Fujian Peoples R China Yunnan Univ Finance & Econ
Sch Stat & Math 237 Longquan Rd Kunming 650224 Yunnan Peoples R China Xiamen Univ
Dept Stat & Data Sci MOE Key Lab Econometr WISESOE 422 Siming South Rd Xiamen 361005 Fujian Peoples R China
The ubiquity of networks in the real world significantly influences dynamic behaviors, with empirical analysis often relying on known network structures to identify network effects. However, inferring latent network s...
详细信息
The ubiquity of networks in the real world significantly influences dynamic behaviors, with empirical analysis often relying on known network structures to identify network effects. However, inferring latent network structures based solely on observed data remains a challenging inverse problem. To this end, we introduce a new method, adaptive Lasso with Multi-directional Signals (ALMS), which transforms the network reconstruction problem into a high-dimensional estimation optimization process. ALMS shrinks the coefficients effectively to different directions by adaptive multi-directional penalties, which is able to reduce the estimation bias. Furthermore, we propose a constrained modification, Constrained adaptive Lasso with Multi-directional Signals (CALMS) to improve both the accuracy and the interpretability of the estimated results. Theoretically, the oracle properties of both the ALMS estimator and the CALMS estimator are established. We demonstrate the excellent finite sample performances of both ALMS and CALMS via comprehensive simulation studies and empirical analysis involving the evolutionary ultimatum game, synchronization model and real experimental data.
Condition assessment for critical infrastructure is a key factor for the wellbeing of the modern human. Especially for the electricity network, specific components such as oil-immersed power transformers need to be mo...
详细信息
Condition assessment for critical infrastructure is a key factor for the wellbeing of the modern human. Especially for the electricity network, specific components such as oil-immersed power transformers need to be monitored for their operating condition. Classic approaches for the condition assessment of oil-immersed power transformers have been proposed in the past, such as the dissolved gases analysis and their respective concentration measurements for insulating oils. However, these approaches cannot always correctly (and in many cases not at all) classify the problems in power transformers. In the last two decades, novel approaches are implemented so as to address this problem, including artificial intelligence with neural networks being one form of algorithm. This paper focuses on the implementation of an adaptive number of layers and neural networks, aiming to increase the accuracy of the operating condition of oil-immersed power transformers. This paper also compares the use of various activation functions and different transfer functions other than the neural network implemented. The comparison incorporates the accuracy and total structure size of the neural network.
Deep learning networks have been trained using first-order-based methods. These methods often converge more quickly when combined with an adaptive step size, but they tend to settle at suboptimal points, especially wh...
详细信息
Deep learning networks have been trained using first-order-based methods. These methods often converge more quickly when combined with an adaptive step size, but they tend to settle at suboptimal points, especially when learning occurs in a large output space. When first-order-based methods are used with a constant step size, they oscillate near the zero-gradient region, which leads to slow convergence. However, these issues are exacerbated under nonconvexity, which can significantly diminish the performance of first-order methods. In this work, we propose a novel Boltzmann Probability Weighted Sine with a Cosine distance-based adaptive Gradient (BSCAGrad) method. The step size in this method is carefully designed to mitigate the issue of slow convergence. Furthermore, it facilitates escape from suboptimal points, enabling the optimization process to progress more efficiently toward local minima. This is achieved by combining a Boltzmann probability-weighted sine function and cosine distance to calculate the step size. The Boltzmann probability-weighted sine function acts when the gradient vanishes and the cooling parameter remains moderate, a condition typically observed near suboptimal points. Moreover, using the sine function on the exponential moving average of the weight parameters leverages geometric information from the data. The cosine distance prevents zero in the step size. Together, these components accelerate convergence, improve stability, and guide the algorithm toward a better optimal solution. A theoretical analysis of the convergence rate under both convexity and nonconvexity is provided to substantiate the findings. The experimental results from language modeling, object detection, machine translation, and image classification tasks on a real-world benchmark dataset, including CIFAR10, CIFAR100, PennTreeBank, PASCALVOC and WMT2014, demonstrate that the proposed step size outperforms traditional baseline methods.
The p-Laplacian problem -del & sdot;((mu + |del u|(p-2))del u) = f is considered, where mu is a given positive number. An anisotropic a posteriori residual-based error estimator is presented. The error estimator i...
详细信息
The p-Laplacian problem -del & sdot;((mu + |del u|(p-2))del u) = f is considered, where mu is a given positive number. An anisotropic a posteriori residual-based error estimator is presented. The error estimator is shown to be equivalent, up to higher order terms, to the error in a quasi-norm. The involved constants being independent of mu, the solution, the mesh size and aspect ratio. An adaptive algorithm is proposed and numerical results are presented when p=3 . From this model problem, we propose a simplified error estimator and use it in the framework of an industrial application, namely a nonlinear Navier-Stokes problem arising from aluminium electrolysis.
Specially tailored skeletal schemes enable cell and face variables linked with a stabilisation and a fine-tuned parameter can provide guaranteed lower eigenvalue bounds for the Laplacian. This paper briefly presents a...
详细信息
Specially tailored skeletal schemes enable cell and face variables linked with a stabilisation and a fine-tuned parameter can provide guaranteed lower eigenvalue bounds for the Laplacian. This paper briefly presents a unified derivation of skeletal higher-order methods from Carstensen, Zhai, and Zhang (2020), Carstensen, Ern, and Puttkammer (2021), and Carstensen, Gr & auml;ss le, and Tran (2024). It suggests a paradigm shift from conditional to unconditional lower eigenvalue bounds. adaptive mesh-refining leads to optimal convergence rates in computational benchmark examples and underlines the superiority of higher-order methods.
In this article, a space-time finite element method (STFEM) for time fractional-order reaction-diffusion equations (TFORDEs) containing fractional derivatives of order alpha is an element of(0,1)\documentclass[12pt]{m...
详细信息
In this article, a space-time finite element method (STFEM) for time fractional-order reaction-diffusion equations (TFORDEs) containing fractional derivatives of order alpha is an element of(0,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in (0,1)$$\end{document} with initial data and boundary value. Utilizing the variational formulation, we establish a space-time finite element framework for addressing the problem. Additionally, we derive a posterior error estimators for both spatial and temporal components, and based on these estimators, we develop an adaptive algorithm. The effectiveness of this adaptive algorithm is ultimately demonstrated through a series of numerical experiments.
In this paper, we study a posteriori error estimates of the fast L1 -2 scheme for time discretization of time fractional parabolic differential equations. To overcome the huge workload caused by the nonlocality of fra...
详细信息
In this paper, we study a posteriori error estimates of the fast L1 -2 scheme for time discretization of time fractional parabolic differential equations. To overcome the huge workload caused by the nonlocality of fractional derivative, a fast algorithm is applied to the construction of the L1 - 2 scheme. Employing the numerical solution obtained by the fast L1 - 2 scheme, a piecewise continuous function approximating the exact solution is constructed. Then, by exploring the error equations, a posteriori error estimates are obtained in different norms, which depend only on the discretization parameters and the data of the problems. Various numerical experiments for the fractional parabolic equations with smooth or nonsmooth exact solutions on different time meshes, including the frequently-used graded mesh, are carried out to verify and complement our theoretical results. Based on the obtained a posteriori error estimates, a time adaptive algorithm is proposed to reduce the computational cost substantially and provides efficient error control for the solution.
Vibration suppression is important for high-precision motion control because positioning accuracy is an essential figure of merit for a servosystem. The paper presents a vibration-suppression method based on improved ...
详细信息
Vibration suppression is important for high-precision motion control because positioning accuracy is an essential figure of merit for a servosystem. The paper presents a vibration-suppression method based on improved adaptive optimal arbitrary-time-delay (IAOAT) input shaping. First, analyzing a conventional optimal-arbitrary-time-delay (OAT) input shaper yields an OAT input shaper with five pulses that completely suppresses two-mode vibrations when the natural frequency and damping ratio are exactly known. Next, embedding an adaptive algorithm in the shaper gives a conventional adaptive optimal arbitrary-time-delay (AOAT) input shaper that handles unknown natural frequencies and damping ratios. Note that the conventional AOAT input shaper contains five pulses that have to be updated, but the delay times of the pulses are prescribed. Then, the IAOAT method extends the symmetry of pulse amplitudes for zero damping ratio to a nonzero case and reduces the number of pulses to three. A recursive-least-squares (RLS) algorithm is devised to update these parameters. This shaper features the smallest number of parameters and high robust performance. Finally, comparisons with other input-shaping methods show the effectiveness and superiority of the developed method over others.
暂无评论