Di-Rong Chen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Di-Rong Chen has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Ding-Xuan Zhou
School of Mathematics and Statistics, University of Sydney, Sydney, Australia
3 shared publications - Jun Fan
Department of Mathematics, Hong Kong Baptist University, Kowloon, Hong Kong, China
3 shared publications - Linhao Song
School of Mathematical Science, Beihang University, Beijing, China
3 shared publications - Hongzhi Tong
School of Statistics, University of International Business and Economics, Beijing 100029, P.R.C.
2 shared publications - Min Han
Beijing University of Technology
2 shared publications - Yunlong Ding
School of Mathematical Science, Beihang University, Beijing 100191, China
2 shared publications
Latest Publications
- 6 August 2024
- Published by World Scientific Pub Co Pte Ltd in International Journal of Modern Physics C
- Vol. 36 (02)
Fractal version of average Fermat distance on some small-world hierarchical networks
Fermat–Wiener index based on topological Wiener index is the total sum of Fermat distance over all the triplets for vertices. In this paper, we construct a class of hierarchical graphs based on hierarchical product generalized from Cartesian product. We study some critical properties of the hierarchical networks by investigating its topological indices. Applying the finite pattern method, we analytically deduce the dominant term of average Fermat distance and obtain its asymptotic formula, which implies small-world property. We finally exhibit a close connection between Fermat–Wiener index and related graph invariants like average geodesic distance, Wiener index and eigenvalues of Laplacian matrix.
- 6 August 2024
- Published by World Scientific Pub Co Pte Ltd in International Journal of Modern Physics C
- Vol. 36 (02)
Fractal version of average Fermat distance on some small-world hierarchical networks
Fermat–Wiener index based on topological Wiener index is the total sum of Fermat distance over all the triplets for vertices. In this paper, we construct a class of hierarchical graphs based on hierarchical product generalized from Cartesian product. We study some critical properties of the hierarchical networks by investigating its topological indices. Applying the finite pattern method, we analytically deduce the dominant term of average Fermat distance and obtain its asymptotic formula, which implies small-world property. We finally exhibit a close connection between Fermat–Wiener index and related graph invariants like average geodesic distance, Wiener index and eigenvalues of Laplacian matrix.
- 22 August 2023
- Published by Springer Nature in Journal of Fourier Analysis and Applications
- Vol. 29 (5) , 1
Correction: Approximation of Nonlinear Functionals Using Deep ReLU Networks
- 22 August 2023
- Published by Springer Nature in Journal of Fourier Analysis and Applications
- Vol. 29 (5) , 1
Correction: Approximation of Nonlinear Functionals Using Deep ReLU Networks
- 28 July 2023
- Published by Springer Nature in Journal of Fourier Analysis and Applications
- Vol. 29 (4) , 1-23
Approximation of Nonlinear Functionals Using Deep ReLU Networks
- 28 July 2023
- Published by Springer Nature in Journal of Fourier Analysis and Applications
- Vol. 29 (4) , 1-23
Approximation of Nonlinear Functionals Using Deep ReLU Networks
- Open Access
- 27 July 2023
- Published by MDPI AG in Mathematics
- Vol. 11 (15) , 3311
Optimization Based Layer-Wise Pruning Threshold Method for Accelerating Convolutional Neural Networks
Among various network compression methods, network pruning has developed rapidly due to its superior compression performance. However, the trivial pruning threshold limits the compression performance of pruning. Most conventional pruning threshold methods are based on well-known hard or soft techniques that rely on time-consuming handcrafted tests or domain experience. To mitigate these issues, we propose a simple yet effective general pruning threshold method from an optimization point of view. Specifically, the pruning threshold problem is formulated as a constrained optimization program that minimizes the size of each layer. More importantly, our pruning threshold method together with conventional pruning works achieves a better performance across various pruning scenarios on many advanced benchmarks. Notably, for the -norm pruning algorithm with VGG-16, our method achieves higher FLOPs reductions without utilizing time-consuming sensibility analysis. The compression ratio boosts from 34% to 53%, which is a huge improvement. Similar experiments with ResNet-56 reveal that, even for compact networks, our method achieves competitive compression performance even without skipping any sensitive layers.- 27 July 2023
- Published by MDPI AG in Mathematics
- Vol. 11 (15) , 3311
Open AccessOptimization Based Layer-Wise Pruning Threshold Method for Accelerating Convolutional Neural Networks
Among various network compression methods, network pruning has developed rapidly due to its superior compression performance. However, the trivial pruning threshold limits the compression performance of pruning. Most conventional pruning threshold methods are based on well-known hard or soft techniques that rely on time-consuming handcrafted tests or domain experience. To mitigate these issues, we propose a simple yet effective general pruning threshold method from an optimization point of view. Specifically, the pruning threshold problem is formulated as a constrained optimization program that minimizes the size of each layer. More importantly, our pruning threshold method together with conventional pruning works achieves a better performance across various pruning scenarios on many advanced benchmarks. Notably, for the -norm pruning algorithm with VGG-16, our method achieves higher FLOPs reductions without utilizing time-consuming sensibility analysis. The compression ratio boosts from 34% to 53%, which is a huge improvement. Similar experiments with ResNet-56 reveal that, even for compact networks, our method achieves competitive compression performance even without skipping any sensitive layers. - Open Access
- 23 April 2023
- Published by MDPI AG in Mathematics
- Vol. 11 (9) , 2001
Learning Bilateral Clipping Parametric Activation for Low-Bit Neural Networks
Among various network compression methods, network quantization has developed rapidly due to its superior compression performance. However, trivial activation quantization schemes limit the compression performance of network quantization. Most conventional activation quantization methods directly utilize the rectified activation functions to quantize models, yet their unbounded outputs generally yield drastic accuracy degradation. To tackle this problem, we propose a comprehensive activation quantization technique namely Bilateral Clipping Parametric Rectified Linear Unit (BCPReLU) as a generalized version of all rectified activation functions, which limits the quantization range more flexibly during training. Specifically, trainable slopes and thresholds are introduced for both positive and negative inputs to find more flexible quantization scales. We theoretically demonstrate that BCPReLU has approximately the same expressive power as the corresponding unbounded version and establish its convergence in low-bit quantization networks. Extensive experiments on a variety of datasets and network architectures demonstrate the effectiveness of our trainable clipping activation function.
- 23 April 2023
- Published by MDPI AG in Mathematics
- Vol. 11 (9) , 2001
Open AccessLearning Bilateral Clipping Parametric Activation for Low-Bit Neural Networks
Among various network compression methods, network quantization has developed rapidly due to its superior compression performance. However, trivial activation quantization schemes limit the compression performance of network quantization. Most conventional activation quantization methods directly utilize the rectified activation functions to quantize models, yet their unbounded outputs generally yield drastic accuracy degradation. To tackle this problem, we propose a comprehensive activation quantization technique namely Bilateral Clipping Parametric Rectified Linear Unit (BCPReLU) as a generalized version of all rectified activation functions, which limits the quantization range more flexibly during training. Specifically, trainable slopes and thresholds are introduced for both positive and negative inputs to find more flexible quantization scales. We theoretically demonstrate that BCPReLU has approximately the same expressive power as the corresponding unbounded version and establish its convergence in low-bit quantization networks. Extensive experiments on a variety of datasets and network architectures demonstrate the effectiveness of our trainable clipping activation function.
- Open Access
Approximation of Nonlinear Functionals Using Deep ReLU Networks
In recent years, functional neural networks have been proposed and studied in order to approximate nonlinear continuous functionals defined on $L^p([-1, 1]^s)$ for integers $s\ge1$ and $1\le p<\infty$. However, their theoretical properties are largely unknown beyond universality of approximation or the existing analysis does not apply to the rectified linear unit (ReLU) activation function. To fill in this void, we investigate here the approximation power of functional deep neural networks associated with the ReLU activation function by constructing a continuous piecewise linear interpolation under a simple triangulation. In addition, we establish rates of approximation of the proposed functional deep ReLU networks under mild regularity conditions. Finally, our study may also shed some light on the understanding of functional data learning algorithms.Open AccessApproximation of Nonlinear Functionals Using Deep ReLU Networks
In recent years, functional neural networks have been proposed and studied in order to approximate nonlinear continuous functionals defined on $L^p([-1, 1]^s)$ for integers $s\ge1$ and $1\le p<\infty$. However, their theoretical properties are largely unknown beyond universality of approximation or the existing analysis does not apply to the rectified linear unit (ReLU) activation function. To fill in this void, we investigate here the approximation power of functional deep neural networks associated with the ReLU activation function by constructing a continuous piecewise linear interpolation under a simple triangulation. In addition, we establish rates of approximation of the proposed functional deep ReLU networks under mild regularity conditions. Finally, our study may also shed some light on the understanding of functional data learning algorithms. - 24 January 2022
- Published by Wiley in Canadian Journal of Statistics
- Vol. 50 (1) , 86-121
Adaptive estimation for functional data: Using a framelet block‐thresholding method
- 24 January 2022
- Published by Wiley in Canadian Journal of Statistics
- Vol. 50 (1) , 86-121
Adaptive estimation for functional data: Using a framelet block‐thresholding method
- 15 May 2019
- Published by Science China Press., Co. Ltd. in SCIENTIA SINICA Mathematica
- Vol. 49 (9) , 1251
Coefficient regularizer based simultaneous regression estimation
It is very important to extract the correlation between two random vectors $X$ and $Y$.The kernel method is used to extract nonlinear correlations.In this paper, the maximum correlation is obtained by minimizing variance ${\rm~Var}[f(X)-g(Y)]$, which is called simultaneous regression, where $f(X)$ and $g(Y)$ are respectively functions in two different reproducing kernel Hilbert spaces.We use regularized empirical variances minimization toconstruct estimates.In order to obtain the estimated function with sparsity,the $l_1$-norm of coefficient is used as the penalty term.We establish a learning rate under some mild conditions.Simultaneous regression is associated with canonical correlation analysis and slice inverse regression, etc.- 15 May 2019
- Published by Science China Press., Co. Ltd. in SCIENTIA SINICA Mathematica
- Vol. 49 (9) , 1251
Coefficient regularizer based simultaneous regression estimation
It is very important to extract the correlation between two random vectors $X$ and $Y$.The kernel method is used to extract nonlinear correlations.In this paper, the maximum correlation is obtained by minimizing variance ${\rm~Var}[f(X)-g(Y)]$, which is called simultaneous regression, where $f(X)$ and $g(Y)$ are respectively functions in two different reproducing kernel Hilbert spaces.We use regularized empirical variances minimization toconstruct estimates.In order to obtain the estimated function with sparsity,the $l_1$-norm of coefficient is used as the penalty term.We establish a learning rate under some mild conditions.Simultaneous regression is associated with canonical correlation analysis and slice inverse regression, etc.- 18 September 2018
- Published by Springer Nature in Bulletin of the Iranian Mathematical Society
- Vol. 45 (3) , 783-789
Improved Approximation Formulas and Inequalities for the Wallis Ratio
- 18 September 2018
- Published by Springer Nature in Bulletin of the Iranian Mathematical Society
- Vol. 45 (3) , 783-789
Improved Approximation Formulas and Inequalities for the Wallis Ratio
- 1 January 2018
- Published by AIP Publishing in AIP Conference Proceedings
- Vol. 1973 (1) , 020027
Semi-supervised hyperspectral image classification with multiscale kernels
In this letter, we propose a novel Laplacian least squares support vector machine in sum space (Lap-LS-SVM-SS) for semi-supervised HIC with multiscale kernels, which can simultaneously deal with the high- and low- frequency components of the target classification function with small and large scale kernels, respectively. The proposed method can solve the multi-class problem of hyperspectral dataset directly. The experimental results on AVIRIS Indian Pines dataset demonstrate that Lap-LS-SVM-SS yields superior classification performance over the conventional semi-supervised classifiers (e.g., LapSVM) in challenging small training labeled samples and a large amount of unlabeled samples.
- 1 January 2018
- Published by AIP Publishing in AIP Conference Proceedings
- Vol. 1973 (1) , 020027
Semi-supervised hyperspectral image classification with multiscale kernels
In this letter, we propose a novel Laplacian least squares support vector machine in sum space (Lap-LS-SVM-SS) for semi-supervised HIC with multiscale kernels, which can simultaneously deal with the high- and low- frequency components of the target classification function with small and large scale kernels, respectively. The proposed method can solve the multi-class problem of hyperspectral dataset directly. The experimental results on AVIRIS Indian Pines dataset demonstrate that Lap-LS-SVM-SS yields superior classification performance over the conventional semi-supervised classifiers (e.g., LapSVM) in challenging small training labeled samples and a large amount of unlabeled samples.