Search

The activation function introduces nonlinearity into convolutional neural network, which greatly promotes the development of computer vision tasks. This paper proposes elastic adaptively parametric compounded units to improve the performance of convolutional neural networks for image recognition. The activation function takes the structural advantages of two mainstream functions as the function's fundamental architecture. The SENet model is embedded in the proposed activation function to adaptively recalibrate the feature mapping weight in each channel, thereby enhancing the fitting capability of the activation function. In addition, the function has an elastic slope in the positive input region by simulating random noise to improve the generalization capability of neural networks. To prevent the generated noise from producing overly large variations during training, a special protection mechanism is adopted. In order to verify the effectiveness of the activation function, this paper uses CIFAR-10 and CIFAR-100 image datasets to conduct comparative experiments of the activation function under the exact same model. Experimental results show that the proposed activation function showed superior performance beyond other functions. [ABSTRACT FROM AUTHOR]

Convolutional neural network-based methods have shown remarkable performance in remote-sensing object detection. However, their deployment on resource-limited embedded devices is hindered by their high computational complexity. Neural network quantization methods have been proven effective in compressing and accelerating CNN models by clipping outlier activations and utilizing low-precision values to represent weights and clipped activations. Nonetheless, the clipping of outlier activations leads to distortion of object local features. Furthermore, the lack of enhanced overall feature mining exacerbates the degradation of detection accuracy. To address the limitations above, we propose an innovative clipping-free quantization method called GradQuant, which mitigates the model’s quantization accuracy loss caused by clipping outlier activations and the lack of overall feature mining. Specifically, a bounded activation function (sigmoid-weighted tanh, SiTanh) is carefully designed to ensure that object features are represented within a limited range without clipping. On the basis of this, an activation substitution training (AST) method is codesigned to prompt models to focus more on nonoutlier object features instead of outlier-like local ones. Extensive experiments on public remote-sensing datasets demonstrate the effectiveness of the GradQuant method compared with other state-of-the-art quantization methods. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

With the flourishing development of wireless communication, further challenges will be introduced by the future demands of emerging applications. However, in the face of more complex communication scenarios, favorable decoding results may not be yielded by conventional channel decoding schemes based on mathematical models. The remarkable contributions of deep neural networks (DNNs) in various fields have garnered widespread recognition, which has ignited our enthusiasm for their application in wireless communication systems. Therefore, a reliable DNN-based decoding scheme designed for wireless communication systems is proposed. This scheme comprises efficient local decoding using linear and nonlinear operations. To be specific, linear operations are carried out on the edges connecting neurons, while nonlinear operations are performed on each neuron. After forward propagation through the DNN, the loss value is estimated based on the output, and backward propagation is employed to update the weights and biases. This process is performed iteratively until a near-optimal message sequence is recovered. Various factors within the DNN are considered in the simulation and the potential impacts of each factor are analyzed. Simulation results indicate that our proposed DNN-based decoding scheme is superior to the conventional hard decision. [ABSTRACT FROM AUTHOR]

Abstract Activation functions are essential in deep learning, and the rectified linear unit (ReLU) is the most widely used activation function to solve the vanishing gradient problem. However, owing to the dying ReLU problem and bias shift effect, deep learning models using ReLU cannot exploit the potential benefits of negative values. Numerous ReLU variants have been proposed to address this issue. In this study, we propose Dynamic Parametric ReLU (DPReLU), which can dynamically control the overall functional shape of ReLU with four learnable parameters. The parameters of DPReLU are determined by training rather than by humans, thereby making the formulation more suitable and flexible for each model and dataset. Furthermore, we propose an appropriate and robust weight initialization method for DPReLU. To evaluate DPReLU and its weight initialization method, we performed two experiments on various image datasets: one using an autoencoder for image generation and the other using the ResNet50 for image classification. The results show that DPReLU and our weight initialization method provide faster convergence and better accuracy than the original ReLU and the previous ReLU variants.

The activation function plays a key role in influencing the performance and training dynamics of neural networks. There are hundreds of activation functions widely used as rectified linear units (ReLUs), but most of them are applied to complex and large neural networks, which often have gradient explosion and vanishing gradient problems. By studying a variety of non-monotonic activation functions, we propose a method to construct a non-monotonic activation function, x·Φ(x), with Φ(x) [0, 1]. With the hardening treatment of Φ(x), we propose an adaptive non-monotonic segmented activation function, called the adaptive quadratic linear unit, abbreviated as AQuLU, which ensures the sparsity of the input data and improves training efficiency. In image classification based on different state-of-the-art neural network architectures, the performance of AQuLUs has significant advantages for more complex and deeper architectures with various activation functions. The ablation experimental study further validates the compatibility and stability of AQuLUs with different depths, complexities, optimizers, learning rates, and batch sizes. We thus demonstrate the high efficiency, robustness, and simplicity of AQuLUs.

The activation function has a critical influence on whether a convolutional neural network in deep learning can converge or not; a proper activation function not only makes the convolutional neural network converge faster but also can reduce the complexity of convolutional neural network architecture and gets the same or better performance. Many activation functions have been proposed; however, various activation functions have advantages, defects, and applicable network architectures. A new activation function called Polynomial Linear Unit (PolyLU) is proposed in this paper to improve some of the shortcomings of the existing activation functions. The PolyLU meets the following basic properties: continuously differentiable, approximate identity near the origin, unbounded for positive inputs, bounded for negative inputs, smooth, monotonic, and zero-centered. There is a polynomial term for the negative inputs and no exponential terms in the PolyLU that reduce the computational complexity of the network. Compared to those common activation functions like Sigmoid, Tanh, ReLU, LeakyReLU, ELU, Mish, and Swish, the experiments show that the PolyLU has improved some network complexity and has better accuracy over MNIST, Kaggle Cats and Dogs, CIFAR-10 and CIFAR-100 datasets. Test by the CIFAR-100 dataset with batch normalization, PolyLU improves by 0.62%, 2.82%, 2.44%, 1.33%, 2.08%, and 4.26% of accuracy than ELU, Swish, Mish, Leaky ReLU, ReLU, and Tanh respectively. Test by the CIFAR-100 dataset without batch normalization, PolyLU improves by 1.24%, 4.39%, 2.12%, 5.43%, 15.51%, and 8.10% of accuracy than ELU, Swish, Mish, Leaky ReLU, ReLU, and Tanh respectively.

A large number of insulators play an important role in insulating and supporting complex power grids, and they are constantly exposed to challenges such as lightning strikes and contamination from the external environment. Only by accurately detecting insulator damage can potential safety hazards and equipment damage due to damaged insulators be avoided in time. Aiming at the problems of low accuracy and large model computation of existing insulator and defect detection algorithms, this paper proposes the insulator and defect detection model YOLO-S (YOLO-Small). First, the PAN structure in Neck uses lightweight convolutional GSConv instead of the standard convolutional Conv, introduces GSbottleneck on the basis of GSConv, and uses the method of OSA to design the cross-level partial network GSCSP module (VoV-GSCSP). The use of VoV-GSCSP instead of the C3 module in Neck reduces the computational effort and maintains the accuracy. Secondly, a new attention module, MaECA (MainECA), is designed based on the ECA attention mechanism to enhance target perception. After that, the SILU function in the SPPF in YOLOv5s is replaced with the Mish function, HardSwish function, and ReLU function, respectively, and the results show that the replacement of the Mish function (MishSPPF) is more effective in preventing the distortion of the image caused by image cropping and scaling, and thus improving the accuracy. Finally, the SIoU loss function is used to replace the original loss function CIoU, which improves the number of transmitted frames per second and the detection accuracy of the pictures. The mAP of YOLO-S is 4.2% higher than that of YOLOv5s, the detection accuracy is improved by 2.1%, and the model parameter computation is reduced by 6.0%, which still gives YOLO-S a higher recognition accuracy when compared with existing detection algorithms under the same conditions.

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.

In this paper, as a geometric approach to the fixed-point theory, we prove new fixed-figure results using the notion of k-ellipse on a metric space. For this purpose, we are inspired by the Caristi type mapping, Kannan type contraction, Chatterjea type contraction and Ćirić type contraction. After that, we give some existence and uniqueness theorems of a fixed k-ellipse. We also support our obtained results with illustrative examples. Finally, we present a new application to the S-Shaped Rectified Linear Activation Unit (SReLU) to show the importance of our theoretical results.

Activation functions are essential components in any neural network model; they play a crucial role in determining the network’s expressive power through their introduced non-linearity. Rectified Linear Unit (ReLU) has been the famous and default choice for most deep neural network models because of its simplicity and ability to tackle the vanishing gradient problem that faces backpropagation optimization. However, ReLU introduces other challenges that hinder its performance; bias shift and dying neurons in the negative region. To address these problems, this paper presents a novel composite monotonic, zero-centered, semi-saturated activation function called Hyperbolic cosine Linearized SquasHing function (HcLSH) with partial gradient-based sparsity HcLSH owns many desirable properties, such as considering the contribution of the negative values of neurons while having a smooth output landscape to enhance the gradient flow during training. Furthermore, the regularization effect resulting from the self-gating property of the positive region of HcLSH reduces the risk of model overfitting and ensures learning more robust expressive representations. An extensive set of experiments and comparisons is conducted that includes four popular image classification datasets, seven deep network architectures, and ten state-of-the-art activation functions. HcLSH exhibited the Top-1 and Top-3 testing accuracy results in 20 and 25 out of 28 conducted experiments, respectively, suppressing the widely used ReLU that achieved 2 and 5, and the reputable Mish that achieved 0 and 5 Top-1 and Top-3 testing accuracy results, respectively. HcLSH attained improvements over ReLU, ranging from 0.2% to 96.4% in different models and datasets. Statistical results demonstrate the significance of the enhanced performance achieved by our proposed HcLSH activation function compared to the competitive activation functions in various datasets and models regarding the testing loss Furthermore, the ablation study further verifies the proposed activation function’s robustness, stability, and adaptability for the different model parameter.

The paper presents in detail a relatively simple implementation method of the hyperbolic tangent function, particularly targeted for FPGAs. The research goal of the proposed method was to examine the usage of the approximation of ordinary or Chebyshev polynomials for the implementation of the function. Several miscellaneous implementation versions have been considered. They differ in the polynomial degree, number of intervals for which the domain of the function is divided, etc. Both floating-point and fixed-point implementations have been presented. An impact on the FPGA resources utilization and calculations time for the implementation versions has also been briefly analyzed. Special attention has been paid to the accuracy of the calculations of the function. It turned out that applying the proposed method, a very high calculations accuracy can be achieved, while simultaneously maintaining reasonable resources utilization and short calculations time. The proposed method can be an effective alternative to other encountered implementation methods such as CORDIC. Additionally, the presented hardware architecture is more versatile and can be easily adapted for the implementation of other mathematical functions.

To explore the extraction of computer image scene and target information, a nonlinear method based on big data technology is proposed. The method can decompose the computer image into a plurality of components when the SAR computer image is processed such as target extraction and computer image compression, which represent different captured image features, respectively. Selecting the most suitable processing method according to the characteristics of different components can greatly improve the performance. Using nonlinear diffusion method, the computer image is decomposed into structural components representing large-scale structural information and texture components representing small-scale detailed information, and the automatic threshold estimation in the diffusion process is studied. The LAIDA criterion is introduced into the automatic threshold solution of nonlinear diffusion-based computer image decomposition to test and evaluate the diffusion process of various diffusion parameter forms. The results show that the experimental outcome of the diffusion decomposition based on automatic threshold estimation is very close on each index, which shows that using automatic threshold estimation, no matter what diffusion index is used, very close results can be obtained. Specifically, for each algorithm, the parameter estimation threshold l for outliers plays an obvious role. The third is the degree of initiative of the estimation process. The larger the L, the larger the outlier, which will lead to a greater extent of the diffusion process, resulting in a continuous decrease in the structural similarity index and compositional correlation. It is proved that the algorithm has strong global search ability, can effectively avoid premature convergence, has fast convergence speed, and good long stability. It can be widely used for optimization of various multimodal functions.

Recent advances in machine learning, particularly with regard to deep learning, help to recognize and classify objects in medical images. In this study, endoscopy images were examined and deep learning method was used to classify healthy and polyp cells. For the proposed system, a database was created from the archives of General Surgery Department Endoscopy Unit in Kutahya Evliya Celebi Training and Research Hospital. The database contains 93 polyps and 216 normal images from 54 archive records. For image multiplexing, a total of 1236 images were obtained by rotating each image 90 degrees around its axis. While 2/3 of the randomly selected data from this obtained data was used for training the model, the rest of the data was reserved for testing. K-fold Cross Validation method was used to reduce the variability of performance results. In this study, 48 different models were created by using different activation and optimization functions to find the best classification model in deep learning. According to the experimental results, it was observed that accuracy of the models depends on the selected parameters; the best model with the accuracy rate of 91% was obtained with 64 neurons in the hidden layer, ReLU activation function and RmsProp optimization method whereas the worst model with the accuracy rate of 76% was obtained with 32 neurons in the hidden layer, Tanh activation and PmsProp optimization functions. Accordingly, classification performance of polyp images can be optimized by utilizing different activation and optimization methods during the design of deep learning models.

As one of the important parts of Neural Network, activation function plays a very important role in model training in Neural Network. In this paper, the status quo, advantages and disadvantages of the existing common activation functions are analysed, and a new activation function is proposed and applied to target detection. To test the performance of the new activation function, this paper compares it with the ReLU activation functions on a variety of Neural Networks and data sets, and not only analyses the performance of the activation function itself but also verifies the effectiveness of the activation function in target detection.

This paper implements an artificial neural network (ANN)-based engine model using the Field Programmable Gate Array (FPGA). The developed (ANN)-based engine model will be used to estimate the engine gas emissions to mitigate the harmful effects of these emissions on human health. Getting reliable and robust FPGA-based ANNs implementations depends on the optimal choice of activation function that will provide minimal area occupation on FPGA. This study introduces, implements, and investigates FPGA-based ANN-based engine models using five different activation functions. These implemented engine models were described using MATLAB/Simulink and hardware description language coder and carried out by Spartan -3E-500.CP132 FPGA platform from Xilinx. The performance of the implemented engine models was investigated in terms of area-efficient implementation and the regression values (R) to build a robust ANN-based engine model.

While non-linear activation functions play vital roles in artificial neural networks, it is generally unclear how the non-linearity can improve the quality of function approximations. In this paper, we present a theoretical framework to rigorously analyze the performance gain of using non-linear activation functions for a class of residual neural networks (ResNets). In particular, we show that when the input features for the ResNet are uniformly chosen and orthogonal to each other, using non-linear activation functions to generate the ResNet output averagely outperforms using linear activation functions, and the performance gain can be explicitly computed. Moreover, we show that when the activation functions are chosen as polynomials with the degree much less than the dimension of the input features, the optimal activation functions can be precisely expressed in the form of Hermite polynomials. This demonstrates the role of Hermite polynomials in function approximations of ResNets.

The world's main source of food supply in agriculture, and due to the increase in population growth, the requirement for food supply is increasing day by day. Rice is the main food crop in many countries, and rice leaf infections are a common hazard to rice yield. So the early detection of rice leaf infection is crucial for rice crop growth and ensuring sufficient supply for a fast-growing population. Conventional manual identification of rice diseases is time-consuming, ineffective, and costly. To overcome this challenge, we can take images of the affected area of the plants and test them with a pre-trained model to identify and classify the rice diseases. Convolutional Neural Networks (CNN) in deep learning extracts the features from the images and diagnoses the diseases efficiently, which addresses the above issues. The main objective of this study is to minimize the loss and the processing time to predict rice diseases. Optimizers play a crucial role in reducing the loss in the neural network. We proposed an Improved Activation and Optimizer Function (IAOF) in the CNN model to minimize the loss and improve the prediction performance and classification accuracy. The output performance of the proposed IAOF-CNN surpasses the other existing methods and has been verified as the best implementation. [ABSTRACT FROM AUTHOR]

This article addresses the multiple μ$$ \mu $$‐stability analysis of n$$ n $$‐dimensional quaternion‐valued neural networks (QVNNs) with unbounded time‐varying delays (UTVD) and two general classes of activation functions (AFs). Firstly, the QVNNs are decomposed into four equivalent real‐valued systems, and based on the geometrical configuration of the AFs, the state space ℍn$$ {\mathbb{H}}^n $$ is divided into 34n$$ {3}^{4n} $$ disjoint regions. Considering the properties of AFs, several sufficient conditions are derived to ensure the coexistence of 34n$$ {3}^{4n} $$ equilibria, out of which 24n$$ {2}^{4n} $$ are locally μ$$ \mu $$‐stable. Moreover, some sufficient conditions for multiple exponential stability, multiple power stability, and multiple log stability are also derived in this article. Finally, two numerical examples are presented. The first example validates the effectiveness of the proposed theoretical results while the second example illustrates the application of QVNNs on the associative memory, which shows that QVNNs have the ability to reliably retrieve true‐color image patterns. [ABSTRACT FROM AUTHOR]

This paper introduces a new spike activation function (SPKAF) or spike membership function for fuzzy adaptive neurons (FAN), developed for decoding spatiotemporal information with spikes, optimizing digital signal processing. A solution with the adaptive network-based fuzzy inference system (ANFIS) method is proposed and compared with that of the FAN-SPKAF model, obtaining very precise simulation results. Stability analysis of systems models is presented. An application to voice recognition using solfeggio syllables in Spanish is performed experimentally, comparing the methods of FAN-step activation function (STEPAF)-SPKAF, Augmented Spiking Neuron Model, and Augmented FAN-STEPAF-SPKAF, achieving very good results. [ABSTRACT FROM AUTHOR]

Detecting and classifying vehicles as objects from images and videos is challenging in appearance-based representation, yet plays a significant role in the substantial real-time applications of Intelligent Transportation Systems (ITSs). The rapid development of Deep Learning (DL) has resulted in the computer-vision community demanding efficient, robust, and outstanding services to be built in various fields. This paper covers a wide range of vehicle detection and classification approaches and the application of these in estimating traffic density, real-time targets, toll management and other areas using DL architectures. Moreover, the paper also presents a detailed analysis of DL techniques, benchmark datasets, and preliminaries. A survey of some vital detection and classification applications, namely, vehicle detection and classification and performance, is conducted, with a detailed investigation of the challenges faced. The paper also addresses the promising technological advancements of the last few years. [ABSTRACT FROM AUTHOR]

The burning of straw is a very destructive process that threatens people's livelihoods and property and causes irreparable environmental damage. It is therefore essential to detect and control the burning of straw. In this study, we analyzed Sentinel-2 data to select the best separation bands based on the response characteristics of clouds, smoke, water bodies, and background (vegetation and bare soil) to the different bands. The selected bands were added to the red, green, and blue bands (RGB) as training sample data. The band that featured the highest detection accuracy, RGB_Band6, was finally selected, having an accuracy of 82.90%. The existing object detection model cannot directly handle multi-band images. This study modified the input layer structure based on the YOLOv5s model to build an object detection network suitable for multi-band remote sensing images. The Squeeze-and-Excitation (SE) network attention mechanism was introduced based on the YOLOv5s model so that the delicate features of smoke were enhanced, and the Convolution + Batch normalization + Leaky ReLU (CBL) module was replaced with the Convolution + Batch normalization + Mish (CBM) module. The accuracy of the model was improved to 75.63%, which was 1.81% better than before. We also discussed the effect of spatial resolution on model detection and where accuracies of 84.18%, 73.13%, and 45.05% for images of 60-, 20-, and 10-m resolution, respectively, were realized. The experimental results demonstrated that the accuracy of the model only sometimes improved with increasing spatial resolution. This study provides a technical reference for the monitoring of straw burning, which is vital for both the control of straw burning and ways to improve ambient air quality. [ABSTRACT FROM AUTHOR]

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. [ABSTRACT FROM AUTHOR]

Accompanied by the fast growth of data and computing power, Deep Learning has developed at a tremendous speed. Other related areas including material science, physics, chemistry, medical science, and engineering also benefit from this well‐proven powerful tool. Activation function is an essential part of Deep Neural Networks, which has attracted a lot of attention, but there are few researches based on nanophotonics problems. In this work, comprehensive research is made on the effectiveness of several widely used activation functions. This research shows that among those fancy activation functions that are investigated, Tanhshrink performs the best, which can predict the spectra with a Root Mean Square Error <0.005 for over 99% of these randomly generated instances. Other traditional activation functions like Tanh and Sigmoid also show excellent outcomes in competition with novel ones. This work shows that although deep learning has already been a powerful tool to solve physics problems, there is still a lot of fundamental work to be done to achieve its maximum potential. [ABSTRACT FROM AUTHOR]

Copyright of Systems Engineering & Electronics is the property of Journal of Systems Engineering & Electronics Editorial Department and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In seismic methods, pore pressure is estimated by converting seismic velocity into pore pressure and calibrating it with pressure results during the well-testing program. This study has been carried out using post-stack seismic data and sonic and density log data of 6 wells in one of the fields in SW Iran. While an optimum number of attributes is selected, the general regression (GRNN) provides higher accuracy than back propagation (BPNN) at the initial prediction stages. However, acoustic impedance (AI) is the most applicable seismic attribute used as root and reverses AI for estimating P-wave and density. Using a set of attributes can train the system to estimate the property. The correlation coefficient of actual and predicted P-wave using an AI seismic attribute has been calculated as 0.74 and the multi-attribute technique as 0.79. Also, density and three attributes reach from 0.57 to 0.60, which shows a better relationship between seismic attributes and density. After determining optimum layers with the principal component analysis (PCA), formation pressure was modeled with the feed forward–back propagation (FFBP-ANN) method. Five information layers, including gamma, VP, AI, density, and overburden pressure, have the most linear convergence with the initial pressure model and are used to modify the ANN model of effective pressure. [ABSTRACT FROM AUTHOR]

Copyright of Computer Measurement & Control is the property of Magazine Agency of Computer Measurement & Control and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Recently, deep learning has achieved remarkable results in fields that require human cognitive ability, learning ability, and reasoning ability. Activation functions are very important because they provide the ability of artificial neural networks to learn complex patterns through nonlinearity. Various activation functions are being studied to solve problems such as vanishing gradients and dying nodes that may occur in the deep learning process. However, it takes a lot of time and effort for researchers to use the existing activation function in their research. Therefore, in this paper, we propose a universal activation function (UA) so that researchers can easily create and apply various activation functions and improve the performance of neural networks. UA can generate new types of activation functions as well as functions like traditional activation functions by properly adjusting three hyperparameters. The famous Convolutional Neural Network (CNN) and benchmark dataset were used to evaluate the experimental performance of the UA proposed in this study. We compared the performance of the artificial neural network to which the traditional activation function is applied and the artificial neural network to which the UA is applied. In addition, we evaluated the performance of the new activation function generated by adjusting the hyperparameters of the UA. The experimental performance evaluation results showed that the classification performance of CNNs improved by up to 5% through the UA, although most of them showed similar performance to the traditional activation function. [ABSTRACT FROM AUTHOR]