Yunfei Zheng

Dr. Yunfei Zheng


Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 July 2020 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
YZYunfei Zheng
Yunfei Zheng
BCBadong Chen
Badong Chen
SWShiyuan Wang
Shiyuan Wang
WWWeiqun Wang
Weiqun Wang
As an effective and efficient discriminative learning method, broad learning system (BLS) has received increasing attention due to its outstanding performance in various regression and classification problems. However, the standard BLS is derived under the minimum mean square error (MMSE) criterion, which is, of course, not always a good choice due to its sensitivity to outliers. To enhance the robustness of BLS, we propose in this work to adopt the maximum correntropy criterion (MCC) to train the output weights, obtaining a correntropy-based BLS (C-BLS). Due to the inherent superiorities of MCC, the proposed C-BLS is expected to achieve excellent robustness to outliers while maintaining the original performance of the standard BLS in the Gaussian or noise-free environment. In addition, three alternative incremental learning algorithms, derived from a weighted regularized least-squares solution rather than pseudoinverse formula, for C-BLS are developed. With the incremental learning algorithms, the system can be updated quickly without the entire retraining process from the beginning when some new samples arrive or the network deems to be expanded. Experiments on various regression and classification data sets are reported to demonstrate the desirable performance of the new methods.

20 October 2020 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
YZYunfei Zheng
Yunfei Zheng
BCBadong Chen
Badong Chen
SWShiyuan Wang
Shiyuan Wang
WWWeiqun Wang
Weiqun Wang
WQWei Qin
Wei Qin
Kernel-based extreme learning machine (KELM), as a natural extension of ELM to kernel learning, has achieved outstanding performance in addressing various regression and classification problems. Compared with the basic ELM, KELM has a better generalization ability owing to no needs of the number of hidden nodes given beforehand and random projection mechanism. Since KELM is derived under the minimum mean square error (MMSE) criterion for the Gaussian assumption of noise, its performance may deteriorate under the non-Gaussian cases, seriously. To improve the robustness of KELM, this article proposes a mixture correntropy-based KELM (MC-KELM), which adopts the recently proposed maximum mixture correntropy criterion as the optimization criterion, instead of using the MMSE criterion. In addition, an online sequential version of MC-KELM (MCOS-KELM) is developed to deal with the case that the data arrive sequentially (one-by-one or chunk-by-chunk). Experimental results on regression and classification data sets are reported to validate the performance superiorities of the new methods.

12 February 2025 in IEEE Transactions on Signal Processing
Verified
Journal Article
Closed access
SCShanli Chen
Shanli Chen
YZYunfei Zheng
Yunfei Zheng
DLDongyuan Lin
Dongyuan Lin
PCPeng Cai
Peng Cai
YXYingying Xiao
Yingying Xiao
SWShiyuan Wang
Shiyuan Wang
Neural network-assisted (NNA) Kalman filters provide an effective solution to addressing the filtering issues involving partially unknown system information by incorporating neural networks to compute the intermediate values influenced by unknown data, such as the Kalman gain in the filtering process. However, whenever there are slight changes in the state-space model (SSM), previously trained networks used in NNA Kalman filters become outdated, necessitating extensive time and data for retraining. Furthermore, obtaining sufficient labeled data for supervised learning is costly, and the effectiveness of unsupervised learning can be inconsistent. To this end, to address the inflexibility of neural network architecture and the scarcity of training data, we propose a model-agnostic meta-learning based neural network-assisted Kalman filter in this paper, called MAML-KalmanNet, by employing a limited amount of labeled data and training rounds to achieve desirable outcomes comparable to the supervised NNA Kalman filters with sufficient training. MAML-KalmanNet utilizes a pre-training approach based on specifically tailored meta-learning, enabling the network to adapt to model changes with minimal data and time without the requirement of retraining. Simultaneously, by fully leveraging the information from the SSM, MAML-KalmanNet eliminates the requirement of a large amount of labeled data to train the meta-learning initialization network. Simulations show that MAML-KalmanNet can mitigate the shortcomings existing in NNA Kalman filters regarding the requirements of abundant training data and sensitive network architecture, while providing real-time state estimation across a range of noise distributions.

1 January 2024 in IEEE/ACM Transactions on Audio, Speech, and Language Processing
Verified
Journal Article
Closed access
QZQiangqiang Zhang
Qiangqiang Zhang
DLDongyuan Lin
Dongyuan Lin
YXYingying Xiao
Yingying Xiao
YZYunfei Zheng
Yunfei Zheng
SWShiyuan Wang
Shiyuan Wang
The conventional active noise control (ANC) system relies on the error-correction learning to design the adaptive controller. However, as the residual noise, the correction error is discarded after each iteration and is still not a perfect input to the human ear. Although many efforts have been devoted to improve the performance by reusing the reference signals, the reuse of the past errors of ANC has been rarely studied in the literature. In this paper, we first design a novel error reused ANC (ErANC) system by introducing the information of the past errors for compensating the gap between the primary noise and estimated output. Then, applying the filtered-x least mean square (FxLMS) algorithm to ErANC for designing the controller, a novel ErFxLMS algorithm is developed to improve the noise reduction performance of FxLMS. However, as the correction errors accumulate over time, the computational and storage burdens of ErFxLMS will grow linearly. To address this issue, a sliding window strategy is applied to the ErFxLMS to curb the size of the stored errors, generating the sliding window ErFxLMS (SW-ErFxLMS) algorithm accordingly. Finally, the performance analysis of mean square error is carried out to validate the universal approximation ability of ErFxLMS. Simulation and experimental results illustrate the effectiveness of the proposed algorithms.

20 March 2023 in Engineering Applications of Artificial Intelligence
Verified
Journal Article
Closed access
YZYunfei Zheng
Yunfei Zheng
SWShiyuan Wang
Shiyuan Wang
BCBadong Chen
Badong Chen

1 March 2025 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
TZTian Zhou
Tian Zhou
SWShiyuan Wang
Shiyuan Wang
Junhui QianJunhui Qian
Junhui Qian
YZYunfei Zheng
Yunfei Zheng
QZQiangqiang Zhang
Qiangqiang Zhang
YXYingying Xiao
Yingying Xiao
Kernel adaptive filters (KAFs) are a class of nonlinear adaptive filters developed in the reproducing kernel Hilbert space, and are particularly suitable for addressing signal processing issues involving data streams and unknown nonlinearities. However, KAFs endure the issue of network structure growth as the number of training samples increases. To this end, a novel structural sparsification method for KAFs, i.e., Gauss Hermite Fourier features (GHFF) method, is first proposed by combining Gauss Hermite quadrature integration rule and Fourier transform. Subsequently, the GHFF method is integrated with the maximum correntropy criterion in filter design, leading to the development of two new adaptive filtering algorithms, i.e., GHFF maximum correntropy (GHFFMC) algorithm and stochastic batch GHFFMC (SB-GHFFMC) algorithm. The proposed GHFFMC and SB-GHFFMC algorithms are expected to exhibit excellent capabilities in characterizing the unknown nonlinear relationships within the data, along with robustness to outliers. Meanwhile, SB-GHFFMC is anticipated to exhibit superior filtering performance in comparison with GHFFMC, as it leverages a general and flexible batch gradient descent method for model optimization. Simulations on nonlinear system identification and time-series prediction of Chua’s circuits confirm the performance superiorities of the proposed algorithms compared to other robust KAFs and RFF-based filters.