Search

Summary: As a significant research issue in control and science field, chaos synchronization has attracted wide attention in recent years. However, it is difficult for traditional control methods to realize synchronization in predefined time and resist external interference effectively. Inspired by the excellent performance of zeroing neural network (ZNN) and the wide application of fuzzy logic system (FLS), a noise‐tolerant fuzzy‐type zeroing neural network (NTFTZNN) with fuzzy time‐varying convergent parameter is proposed for the synchronization of chaotic systems in this paper. Notably the fuzzy parameter generated from FLS combined with traditional convergent parameter embedded into this NTFTZNN can adjust the convergence rate according to the synchronization errors. For the sake of emphasizing the advantages of NTFTZNN model, other three sets of contrast models (FTZNN, VPZNN, and PTZNN) are constructed for the purpose of comparison. Besides, the predefined‐time convergence and noise‐tolerant ability of NTFTZNN model are distinctly demonstrated by detailed theoretical analysis. Furthermore, synchronization simulation experiments including two chaotic systems with different dimensions are provided to verify the related mathematical theories. Finally, the schematic of NTFTZNN model for chaos synchronization is accomplished completely through Simulink, further accentuating its effectiveness and potentials in practical applications. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) are finding increasing use as tools to model and solve problems in almost every discipline in today's world. The successful implementation of ANNs in software—particularly in the fields of deep learning and machine learning—has spiked an interest in designing hardware architectures that are custom-made to implement ANNs. Several categories of ANNs exist. The two-layer bidirectional associative memory (BAM) is a particular class of hetero-associative memory networks that is extremely efficient and exhibits good performance for storing and retrieving pattern pairs. The memristor is a novel hardware element that is well-suited to modelling neural synapses because it exhibits tunable resistance. In this work, in order to create a device that can perform Braille–Latin conversion, we have implemented a circuit realization of a BAM neural network. The implemented hardware BAM uses a memristor crossbar array for modelling neural synapses and a neuron circuit comprising an I-to-V converter (resistor), voltage comparator, D flip-flop, and inverter. The efficiency of the implemented hardware BAM was tested initially using 2 × 2 and 3 × 3 patterns. Upon successfully verifying the ability of the implemented BAM to store and retrieve simple pattern pairs, it was trained for a pattern-recognition application, namely mapping Braille alphabets to their Latin counterparts and vice versa. The performance of the implemented BAM network is robust even on the introduction of noise. The application can recognize the input patterns with accuracies of 100% in either direction when tested with up to 30% noise. [ABSTRACT FROM AUTHOR]

In recent years, neural networks have become a common practice in academia for handling complex problems. Numerous studies have indicated that complex problems can generally be formulated as a single or a set of time-varying equations. Dynamic neural networks, as powerful tools for processing time-varying problems, play an essential role in their online solution. This paper reviews recent advances in real-valued, complex-valued, and noise-tolerant dynamic neural networks for solving various time-varying problems, discusses the finite-time convergence, fixed/varying parameters, and noise tolerance properties of dynamic neural network models. This review is highly relevant for researchers and novices interested in using dynamic neural networks to solve time-varying problems.

Abstract Background Retinal vessel segmentation benefits significantly from deep learning. Its performance relies on sufficient training images with accurate ground-truth segmentation, which are usually manually annotated in the form of binary pixel-wise label maps. Manually annotated ground-truth label maps, more or less, contain errors for part of the pixels. Due to the thin structure of retina vessels, such errors are more frequent and serious in manual annotations, which negatively affect deep learning performance. Methods In this paper, we develop a new method to automatically and iteratively identify and correct such noisy segmentation labels in the process of network training. We consider historical predicted label maps of network-in-training from different epochs and jointly use them to self-supervise the predicted labels during training and dynamically correct the supervised labels with noises. Results We conducted experiments on the three datasets of DRIVE, STARE and CHASE-DB1 with synthetic noises, pseudo-labeled noises, and manually labeled noises. For synthetic noise, the proposed method corrects the original noisy label maps to a more accurate label map by 4.0– $$9.8\%$$ 9.8 % on $$F_1$$ F 1 and 10.7– $$16.8\%$$ 16.8 % on PR on three testing datasets. For the other two types of noise, the method could also improve the label map quality. Conclusions Experiment results verified that the proposed method could achieve better retinal image segmentation performance than many existing methods by simultaneously correcting the noise in the initial label map.

Since the beginning of the COVID-19 pandemic, the demand for unmanned aerial vehicles (UAVs) has surged owing to an increasing requirement of remote, noncontact, and technologically advanced interactions. However, with the increased demand for drones across a wide range of fields, their malicious use has also increased. Therefore, an anti-UAV system is required to detect unauthorized drone use. In this study, we propose a radio frequency (RF) based solution that uses 15 drone controller signals. The proposed method can solve the problems associated with the RF based detection method, which has poor classification accuracy when the distance between the controller and antenna increases or the signal-to-noise ratio (SNR) decreases owing to the presence of a large amount of noise. For the experiment, we changed the SNR of the controller signal by adding white Gaussian noise to SNRs of −15 to 15 dB at 5 dB intervals. A power-based spectrogram image with an applied threshold value was used for convolution neural network training. The proposed model achieved 98% accuracy at an SNR of −15 dB and 99.17% accuracy in the classification of 105 classes with 15 drone controllers within 7 SNR regions. From these results, it was confirmed that the proposed method is both noise-tolerant and scalable.

Multilayer dynamic system is widely used in industry and other fields. Different from common systems, multilayer dynamic system has complex structure leading to challenges for research. In this paper, we study zeroing neural network(ZNN) models for a class of multilayer dynamic system(MLDS). In the case of continuous time, continuous ZNN models for continuous MLDS (MLDS-linear-ZNN, MLDS-nonlinear-ZNN and MLDS-noise-tolerant-ZNN) are proposed based on ZNN design method with theoretical analysis. In the discrete case concurrently, discrete ZNN models (MLDS-linear- m DZNN, MLDS-nonlinear- m DZNN and MLDS-noise-tolerant- m DZNN) with m -step ZeaD formula, a new Zhang et al. discretization formula presented in previous paper, are put forward and corresponding discrete algorithms are obtained. Finally, numerical experiments are carried out to verify the superiority and maneuverability of ZNN models for MLDS proposed in this paper. [ABSTRACT FROM AUTHOR]

Background: Retinal vessel segmentation benefits significantly from deep learning. Its performance relies on sufficient training images with accurate ground-truth segmentation, which are usually manually annotated in the form of binary pixel-wise label maps. Manually annotated ground-truth label maps, more or less, contain errors for part of the pixels. Due to the thin structure of retina vessels, such errors are more frequent and serious in manual annotations, which negatively affect deep learning performance. Methods: In this paper, we develop a new method to automatically and iteratively identify and correct such noisy segmentation labels in the process of network training. We consider historical predicted label maps of network-in-training from different epochs and jointly use them to self-supervise the predicted labels during training and dynamically correct the supervised labels with noises. Results: We conducted experiments on the three datasets of DRIVE, STARE and CHASE-DB1 with synthetic noises, pseudo-labeled noises, and manually labeled noises. For synthetic noise, the proposed method corrects the original noisy label maps to a more accurate label map by 4.0– 9.8 % on F 1 and 10.7– 16.8 % on PR on three testing datasets. For the other two types of noise, the method could also improve the label map quality. Conclusions: Experiment results verified that the proposed method could achieve better retinal image segmentation performance than many existing methods by simultaneously correcting the noise in the initial label map. [ABSTRACT FROM AUTHOR]

In the previous work, a finite-time zeroing neural network (ZNN) has been established to find the accurate solution of Lyapunov equation in the presence of no noises. In order to further improve the convergence speed of ZNN and suppress various noises encountered in real applications, in this paper, two robust nonlinear zeroing neural networks (RNZNNs) are designed by adding two novel nonlinear activation functions (AFs) for finding the solution of the Lyapunov equation in the presence of various noises. Unlike the previous ZNN activated by known AFs (e.g., linear activation function, bipolar sigmoid activation function, and power activation function), the proposed two RNZNN models possess predefined-time convergence (instead of finite-time convergence) even in the presence of various noises. The greatest advantage of the predefined-time convergence is independent to initial states of a dynamic system, which is much superior to the finite-time convergence related to initial states, and tremendously modifies the convergence performance. In addition, the predefined-time convergence of the RNZNN models for solving the Lyapunov equation are mathematically proved in detail under various external noises. The simulation comparisons further verify the superiority of the proposed RNZNN models for finding the solution of the Lyapunov equation. [ABSTRACT FROM AUTHOR]

Noise immunity is the foremost issue in high-speed domino circuits. In general, better noise immunity is achieved at the cost of speed and power degradation. In this paper, pseudo-dynamic keeper design is proposed to reduce the delay and power with improved noise immunity for domino circuits. The proposed technique is able to achieve reduced delay, power consumption, and better noise immunity by using always ON keeper. The simulation results show that the proposed technique exhibits 41%, 39%, and 19% delay reduction when compared with the low power dynamic circuit for two-input OR gate, two-input EX-OR gate, and 4:1 multiplexer. The proposed logic also performs better as compared to a low power dynamic circuit with 24%, 21%, and 14% reduction in power-delay product for two-input OR gate, twoinput EX-OR gate, and four input MUX, respectively. The unity noise gain is also improved as compared to all other existing methods. [ABSTRACT FROM AUTHOR]

Recently, a new formula with noise-tolerant capability has been designed by Zhanget al., and the resultant Zhang dynamics (ZD) models have been developed to solve different types of time-varying problems. Based on previous research, this paper presents and investigates the application of such a noise-tolerant ZD design formula to kinematic control of redundant robot manipulators via time-varying nonlinear equations solving. Specifically, by exploiting this ZD design formula to solve the system of time-varying nonlinear kinematic equations involved in robot control, the redundancy-resolution scheme is established. Such a scheme contains the proportional, integral, and derivative information of the desired Cartesian path of the robot end-effector and can thus be viewed as a nonlinear proportional-integral-derivative controller for redundant robot manipulators. Then, theoretical results are given to show that the redundancy-resolution scheme has the capability of noise suppressing. Simulation results based on a four-link planar robot manipulator and a PA10 robot manipulator with zero, constant, and bounded time-varying noises further substantiate the efficacy and superiority of the redundancy-resolution scheme, and show the application prospect of the noise-tolerant ZD design formula. [ABSTRACT FROM AUTHOR]

Accurate and sufficient location information is the prerequisite for most wireless sensor networks (WSNs) applications. Existing range-based localization approaches often suffer from incomplete and corrupted range measurements. Recently, some matrix completion-based localization approaches have been proposed, which only take into account Gaussian noise and outlier noise when modeling the range measurements. However, in some real-world applications, the inevitable structural noise usually degrades the localization accuracy and prevents the outlier recognition drastically. To address these challenges, we propose a noise-tolerant localization via multi-norms regularized matrix completion (LMRMC) approach in this paper. Leveraging the intrinsic low-rank property of euclidean distance matrix (EDM), the reconstruction problem of true underlying EDM is formulated as a multi-norms regularized matrix completion model, where the outlier noise and structural noise are explicitly sifted by $L_1$-norm and L_{1,2}-norm, respectively, while the Gaussian noise is implicitly smoothed by employing the well-known alternating direction method of multiplier optimization method. To the best of our knowledge, this is the first scheme being able to efficiently recover the unknown range measurements under the coexistence of Gaussian noise, outlier noise, and structural noise. Extensive experiments validate the superiority of our proposed LMRMC approach, outperforming the state-of-the-art localization approaches with regard to the localization accuracy. Besides, LMRMC can also achieve an accurate detection of both outlier noise and structural noise, making it promising for further nodes fault diagnosis and topology control in WSNs. [ABSTRACT FROM AUTHOR]

• A view- and altitude-invariant drone object detector. • A binary MF integrating multiple attributes for calculating graphlet's hash codes. • A million-scale high-resolution drone image set for performance evaluation. Drone, a.k.a. Unmanned aerial vehicle (UAV), has been pervasively applied in geological hazard monitoring, smart agriculture, and urban planning in the past decade. In this work, we fuse multiple attributes into a noise-tolerant hashing framework that can detect objects from drone pictures extremely fast. Our method can intrinsically and flexibly encode various topological structures from each target object, based on which multi-scale objects can be discovered in a view- and altitude-invariant way. Moreover, by leveraging l F and l 1 norms collaboratively, the calculated hash codes are robust to low quality drone pictures and noisy semantic labels. More specifically, for each drone-borne picture, we extract visually/semantically salient object parts inside it. To characterize their topological structure, we construct a graphlet by linking the spatially adjacent object patches into a small graph. Subsequently, a binary matrix factorization (MF) is designed to hierarchically exploit the semantics of these graphlets, wherein three attributes: i) deep binary hash codes learning, ii) contaminated pictures/labels denoising, and iii) adaptive data graph updating are seamlessly incorporated. Such multi-attribute binary MF can be solved iteratively, and in turn each graphlet is transformed into the binary hash codes. Finally, the hash codes corresponding to graphlets within each drone photo are utilized for ranking-based object discovery. Comprehensive experiments on the DAC-SDC, MOHR, and our self-compiled data set have demonstrated the competitively speed and accuracy of our method. As a byproduct, we employ an elaborately-designed FPGA architecture to calculate our hash codes. On average, a 57 frames per second (fps) object detection speed is achieved on 4K drone videos (without temporal modeling). [ABSTRACT FROM AUTHOR]

In this paper, a true-single-phase clock latching based noise-tolerant (TSPCL-NT) design for dynamic CMOS circuits is proposed. A TSPCL-NT dynamic circuit can isolate and filter noise before the noise enters into the dynamic circuit. Therefore, it cannot only greatly enhance the noise tolerance of dynamic circuits but also release the signal contention between the feedback keeper and the pull-down network effectively. As a result, noise tolerance of dynamic circuits can be improved with lower sacrifice in power consumption and operating speed. In the 16-bit TSPCL-NT Manchester adder, the average noise threshold energy can be enhanced by 3.41 times. In the meanwhile, the power-delay product can be improved by 5.92% as compared with the state-of-the art 16-bit XOR-NT Manchester adder design under TSMC 90 nm CMOS process. Copyright © 2014 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]

Abstract: As the size of CMOS devices is scaled down to lower the power consumption and space occupied on the chip to the nano-scale, unfortunately, noise is not reduced accordingly. As a result, interference due to noise can significantly affect circuit performance and operation. Since noises are random and dynamic in nature, probabilistic noise-tolerant approaches are more desirable to handle this problem. However, trade-offs between hardware complexity and noise-tolerance are severe design challenges in the probabilistic-based noise-tolerant approaches. In this paper, we proposed a cost-effective common-feedback probabilistic-based noise-tolerant VLSI circuit based on Markov random field (MRF) theory. We proposed a common latch feedback method to lower the hardware complexity. To further enhance the noise-tolerant ability, the common latch feedback technique is combined with Schmitt trigger. To demonstrate the proof-of-concept design, a 16-bit carry-lookahead adder was implemented in the TSMC 90nm CMOS process technology. As compared with the state-of-art master-and-slave MRF design, the experimental results show that not only the transistor count can be saved by 20%, the noise-tolerant performance can also be enhanced from 18.1dB to 24.2dB in the proposed common feedback MRF design. [Copyright &y& Elsevier]

Summary: Objective: This paper presents a noise-tolerant electrocardiogram (ECG) beat classification method based on higher order statistics (HOS) of subband components. Methods and material: Five levels of discrete wavelet transform (DWT) were applied to decompose the signal into six subband components. Higher order statistics proceeded to calculate four sets of HOS features from the three midband components, which together with three RR interval-related features constructed the primary feature set. A feature selection algorithm based on correlation coefficient and Fisher discriminality was then exploited to eliminate redundant features from the primary feature set. A feedforward backpropagation neural network (FFBNN) was employed as the classifier. Two sample selection strategies and four categories of noise artifacts were utilized to justify the capacity of the method. Results: More than 97.5% discrimination rate was achieved, no matter which of the two sampling selection strategies was used. By using the feature selection method, the feature dimension can be readily reduced from 30 to 18 with negligible decrease in accuracy. Compared with other method in the literature, the proposed method improves the sensitivities of most beat types, resulting in an elevated average accuracy. The proposed method is tolerant to environmental noises; as high as 91% accuracies were retained even when contaminated with serious noises, 10dB signal-to-noise ration (SNR), of different kinds. Conclusion: The results demonstrate the effectiveness and noise-tolerant capacities of the proposed method in ECG beat classification. [Copyright &y& Elsevier]

Recently, Xiao et al. presented a new noise-tolerant zeroing neural network (NNTZNN) model that can achieve fixed-time convergence even in the presence of some noise to solve the time-varying Sylvester equation. In this brief paper, we extend it to more general case through introducing a power parameter k while retaining its fixed-time convergence and noise-tolerant performance. The new proposed model is here named as the accelerating noise-tolerant zeroing neural network (ANTZNN) model since its fixed convergence time can be shorter than that of the NNTZNN model in certain model parameter ranges. These parameter ranges and the convergence of the ANTZNN model are theoretically analyzed in detail. Numerical experiments are performed to confirm the theoretical results, including the numerical comparisons with the known NNTZNN model under different parameter settings. Furthermore, the ANTZNN model is also applied to the control of robot manipulator, thus showing the applicability of the proposed ANTZNN model. [ABSTRACT FROM AUTHOR]

One important strategy in semi-supervised learning is to utilize the predicted pseudo labels of unlabeled data to relieve the overdependence on the ground truth of supervised learning algorithms. However, the performance of such kinds of semi-supervised methods heavily relies on the quality of pseudo labels. To address this issue, a robust semi-supervised classification method, named data augmented online extreme learning machines (ELMs) with deep features (DF-DAELM) is proposed. This method firstly extracts features and infers labels for unlabeled data through self-training. Then, with the learned features and inferred labels, two noise-robust shallow classifiers based on data augmentation (i.e., SLI-OELM and CR-OELM) are proposed to eliminate the adverse effects of noises on classifier training. Specifically, inspired by label smoothing, a data augmented method, SLI-OELM is designed based on stochastic linear interpolation to improve the robustness of classifiers based on ELMs. Furthermore, based on the smoothing assumption, the proposed CR-OELM utilizes an ℓ 2 -norm consistency regularization term to implicitly weight noisy samples. Comprehensive experiments demonstrate that DF-DAELM achieves competitive or even better performance on CIFAR-10/100 and SVHN over the related state-of-the-art methods. Meanwhile, for the proposed classifiers, experimental results on the MNIST dataset with different noise levels and sample scales demonstrate their superior performance, especially when the sample scale is small (≤ 20 K) and the noise is strong (40 % ∼ 80 %). [ABSTRACT FROM AUTHOR]

In this paper, a novel Taylor-type difference rule with O ( τ 4 ) pattern error is provided for the first-order derivative approximation. Then, a high accuracy noise-tolerant five-step discrete-time zeroing neural network (ZNN) (termed as FDNTZNN model) is proposed to solve the time-varying matrix inversion problem in real-time. In addition, to obtain the derivative value of time-varying variables in real-world applications, the backward-difference rule is exploited to develop the FD-NTZNN model when the derivative information is unknown (FD-NTZNN-U). Theoretical analysis demonstrates that the proposed FD-NTZNN models have the properties of 0 − stability, consistency and convergence. For comparative analysis, the classical Euler-type discrete-time ZNN model (EDZNN), five-step Taylor-type discrete-time ZNN model (FDZNN) and Euler-type discrete-time noise-tolerant ZNN (NTZNN) model (ED-NTZNN) are reconsidered. Ultimately, two illustrative numerical simulations and an application example to motion generation of manipulator are simulated to substantiate the feasibility and effectiveness of the proposed FD-NTZNN model and FD-NTZNN-U model for online time-varying matrix inversion in the presence of different types of noise. • Two models are developed for time varying matrix inversion with different noises. • Theoretical analyses show the proposed models have higher accuracy than ZNN model. • Two models have superior stability and robustness under different noises. [ABSTRACT FROM AUTHOR]

• Minimal threshold increases of the H 2 neurons in the inferior colliculus were comparable to increase in bat's call intensity observed in 70 dB SPL white noise condition. • The H 2 neurons have higher minimal thresholds. • The H 2 neurons have sharper frequency tuning curves. • Background white noise can sharpen intensity tuning of inferior collicular neurons. • The H 2 neurons have consistent best amplitude spikes and sharper intensity tuning in background noise condition than in silence. Although acoustic communication is inevitably influenced by noise, behaviorally relevant sounds are perceived reliably. The noise-tolerant and -invariant responses of auditory neurons are thought to be the underlying mechanism. So, it is reasonable to speculate that neurons with best frequency tuned to behaviorally relevant sounds will play important role in noise-tolerant perception. Echolocating bats live in groups and emit multiple harmonic signals and analyze the returning echoes to extract information about the target features, making them prone to deal with noise in their natural habitat. The echolocation signal of Hipposideros pratti usually contains 3–4 harmonics (H 1 H 4), the second harmonic has the highest amplitude and is thought to play an essential role during echolocation behavior. Therefore, it is reasonable to propose that neurons tuned to the H 2 , named the H 2 neurons, can be more noise-tolerant to background noise. Taking advantage of bat's stereotypical echolocation signal and single-cell recording, our present study showed that the minimal threshold increases (12.2 dB) of H 2 neurons in the auditory midbrain were comparable to increase in bat's call intensity (14.2 dB) observed in 70 dB SPL white noise condition, indicating that the H 2 neurons could work as background noise monitor. The H 2 neurons had higher minimal thresholds and sharper frequency tuning, which enabled them to be more tolerant to background noise. Furthermore, the H 2 neurons had consistent best amplitude spikes and sharper intensity tuning in background white noise condition than in silence. Taken together, these results suggest that the H 2 neurons might account for noise-tolerant perception of behaviorally relevant sounds. [ABSTRACT FROM AUTHOR]