Balanced Loss Function Research Articles

Research on the Integration of Digital Media and Oil Painting Teaching in Colleges and Universities in the Era of Artificial Intelligence

Abstract Under the impetus of new concepts and new thinking, digital media art is also born. This paper discusses the integration of digital media in oil painting teaching in colleges and universities and constructs a resource library of painting images created by college and university students. The original RGB image is converted to HSV mode, and the college oil painting images are classified into styles based on color entropy. Drawing on the category balanced intersectionality loss function commonly used in edge extraction networks and designing an objective evaluation index based on the Adain network model for comparing the stylistic categorization ability of oil painting image translation networks. Finally, a generalization experiment was conducted on oil painting teaching painting images in colleges and universities through the Gallerix dataset. The results show that in the 2566-dimensional vector data distribution extracted from the oil painting resource base of art majors in H colleges and universities, for example, with dc = 0.05 and dc = 0.1, both sets of parameters find the correct clustering centers, which greatly enhances the efficiency of oil painting creation by college and university students. The method can promote the further development of artificial intelligence technology in the field of art creation and evaluation.

Unsupervised image enhancement under non-uniform illumination based on paired CNNs

This paper presents two CNN-based systems for unsupervised image enhancement under non-uniform illumination. The core of the systems is constituted by the difference of a pair of CNNs. Each CNN is composed of two convolutional layers of neurons with exponential activation function and logarithmic activation function. A weighted sum of the non-reference loss functions is used to train the paired CNNs. It includes an entropy enhancement function and a Bézier loss function to ensure global and local enhancement complementarily. It also includes a white balance loss function to remove color cast in raw images, and a gradient improvement loss function to compensate for the high frequency degradation . In addition, it includes an SSIM (structural similarity index) loss functions to ensure image fidelity. In addition to the basic system, CNNOD, an augmented version called CNNOD+ is developed, which features an information fusion/combination module with a power-law network for gamma correction. The experimental results on two benchmark datasets are discussed to demonstrate that the proposed systems outperform the state-of-the-art methods in terms of enhancement quality, model complexity, and convergence efficiency.

Bayesian and E-Bayesian estimation based on constant-stress partially accelerated life testing for inverted Topp–Leone distribution

Abstract Accelerated or partially accelerated life tests are particularly significant in life testing experiments since they save time and cost. Partially accelerated life tests are carried out when the data from accelerated life testing cannot be extrapolated to usual conditions. The constant-stress partially accelerated life test is proposed in this study based on a Type-II censoring scheme and supposing that the lifetimes of units at usual conditions follow the inverted Topp–Leone distribution. The Bayes and E-Bayes estimators of the distribution parameter and the acceleration factor are derived. The balanced squared error loss function, which is a symmetric loss function, and the balanced linear exponential loss function, which is an asymmetric loss function, are considered for obtaining the Bayes and E-Bayes estimators. Based on informative gamma priors and uniform hyper-prior distributions, the estimators are obtained. Finally, the performance of the proposed Bayes and E-Bayes estimates is evaluated through a simulation study and an application using real datasets.

Investigation of Exponential Distribution Utilizing Randomly Censored Data under Balanced Loss Functions and Its Application to Clinical Data

In this research, random censoring is employed as a methodology for parameter estimation within the context of an exponential distribution. These parameter estimations are conducted using both the Bayesian and maximum likelihood approaches. In the Bayesian framework, Lindley’s approximation method is applied to derive estimates, which are subsequently assessed under three distinct balanced loss functions. To gauge the efficacy of different estimation techniques, simulation-based investigations are conducted. Additionally, a real-world data analysis is executed to illustrate the practical applicability of these methodologies. The findings consistently underscore the superiority of Bayesian parameter estimates in comparison with their maximum likelihood counterparts across all analyzed methodologies.

An improvement of James-Stein estimator under the balanced loss function

An improvement of James-Stein estimator under the balanced loss function

Automatic segmentation of brain glioma based on XY-Net.

Glioma is a malignant primary brain tumor, which can easily lead to death if it is not detected in time. Magnetic resonance imaging is the most commonly used technique to diagnose gliomas, and precise outlining of tumor areas from magnetic resonance images (MRIs) is an important aid to physicians in understanding the patient's condition and formulating treatment plans. However, relying on radiologists to manually depict tumors is a tedious and laborious task, so it is clinically important to investigate an automated method for outlining glioma regions in MRIs. To liberate radiologists from the heavy task of outlining tumors, we propose a fully convolutional network, XY-Net, based on the most popular U-Net symmetric encoder-decoder structure to perform automatic segmentation of gliomas. We construct two symmetric sub-encoders for XY-Net and build interconnected X-shaped feature map transmission paths between the sub-encoders, while maintaining the feature map concatenation between each sub-encoder and the decoder. Moreover, a loss function composed of the balanced cross-entropy loss function and the dice loss function is used in the training task of XY-Net to solve the class unevenness problem of the medical image segmentation task. The experimental results show that the proposed XY-Net has a 2.16% improvement in dice coefficient (DC) compared to the network model with a single encoder structure, and compare with some state-of-the-art image segmentation methods, XY-Net achieves the best performance. The DC, HD, recall, and precision of our method on the test set are 74.49%, 10.89mm, 78.06%, and 76.30%, respectively. The combination of sub-encoders and cross-transmission paths enables the model to perform better; based on this combination, the XY-Net achieves an end-to-end automatic segmentation of gliomas on 2D slices of MRIs, which can play a certain auxiliary role for doctors in grasping the state of illness.

FL-IIDS: A novel federated learning-based incremental intrusion detection system

With the advantage of analyzing data of multiple work sites comprehensively while ensuring data privacy, federated learning-based intrusion detection systems (IDS) are emerging as a distributed intrusion detection paradigm. Most of these IDS are assumed to work on static data. However, in the actual network environment, the practice of setting data as static will lead to the phenomenon known as catastrophic forgetting, where old classes that have already appeared would be forgotten. In this paper, we propose a novel IDS framework called FL-IIDS to effectively address the catastrophic forgetting problem. Firstly, a new loss function is synthetically designed for local model training. With the new function, the class gradient balance loss function assigns different learning weights to data of the new and old classes so that the learning rate of the new classes would decrease and the memory of the overall old classes would be deepened. Moreover, the sample label smoothing loss function leverages the knowledge distillation method to enhance the local model memory for every specific class of old classes. Secondly, the relay client fusing sample reconstruction is employed to mitigate the spread of catastrophic forgetting globally without compromising data privacy. Extensive experimental results on the UNSW-NB15 dataset and the CICIDS2018 dataset show that our proposed framework improves the memory capability for old classes substantially without affecting the detection effectiveness of the IDS for new classes.

EDSV-Net: An efficient defect segmentation network based on visual attention and visual perception

In industrial production, surface defect detection algorithms based on convolutional neural networks have been widely studied to improve production quality. However, for practical applications, there are still many issues to be solved, such as the complexity and diversity of defect categories, the difficulty of obtaining defect samples, and the difficulty of existing algorithms in accurately segmenting defects. To solve these issues, we present an effective defect segmentation network based on visual attention and visual perception termed EDSV-Net. Specifically, we use ResNet18 as the backbone network in EDSV-Net. Then a multi-scale feature extraction (MSFE) module is introduced to enhance the scale invariance of high-level features and the diversity of contextual features. In addition, a spatial attention (SA) model combined with a channel attention (CA) model is applied to low level features and MSFE features, respectively, to extract more effective spatial and semantic information. Moreover, a depthwise separable convolution is introduced to reduce the network complexity. Finally, due to the issues of existing defect detection algorithms ignoring structural similarity and defects being difficult to obtain, we design a balanced defect and structural measure loss function. Meanwhile, we propose a structural similarity measure, which combines the pixel similarity for evaluation. EDSV-Net only requires no more than 60 random abnormal samples to obtain accurate segmentation results and the real-time performance meets the requirements of actual industrial production. Based on three challenging real-world defect datasets, the results of the evaluation demonstrate that EDSV-Net outperforms seven state-of-the-art methods on accuracy and real-time performance.

Utilizing Empirical Bayes Estimation to Assess Reliability in Inverted Exponentiated Rayleigh Distribution with Progressive Hybrid Censored Medical Data

This study addresses the issue of estimating the shape parameter of the inverted exponentiated Rayleigh distribution, along with the assessment of reliability and failure rate, by utilizing Type-I progressive hybrid censored data. The study explores the estimators based on maximum likelihood, Bayes, and empirical Bayes methodologies. Additionally, the study focuses on the development of Bayes and empirical Bayes estimators with balanced loss functions. A concrete example based on actual data from the field of medicine is used to illustrate the theoretical insights provided in this study. Monte Carlo simulations are employed to conduct numerical comparisons and evaluate the performance and accuracy of the estimation methods.

Digital printing image generation method based on style transfer

Digital printing has been widely used in textile printing production. In the process of designing digital printing patterns, an image generative model is needed to assist in obtaining more diversified patterns. However, the current model involves large storage space and high computing cost, which affects the promotion of digital printing customized production. To solve the problem, this article proposes a digital printing image generation method based on style transfer. Firstly, a style transfer method based on exact feature distribution matching is constructed to realize the accurate matching from image content to style features. And a balanced loss function is used to enhance the universality of the proposed method. Furthermore, knowledge distillation is introduced to compress the method proposed to reduce the hardware requirements when processing high-resolution digital printing images. Finally, a segmented training strategy is proposed to solve the performance degradation caused by model compression. The experimental results show that when processing images with a resolution of 3000 × 3000, the storage capacity of the model is only 2.68 MB and only 0.20 TFLOPs is required. The maximum processing resolution is more than 8K. The pattern obtained by this model is of high quality and can meet the needs of digital printing production.

A landslide susceptibility assessment method based on auto-encoder improved deep belief network

Abstract The landslide susceptibility assessment is an essential part of landslide disaster risk identification and prevention. However, the binarization of the hidden layer limits the parameterization ability of the conditional probability of visible layer, making the training process of restricted Boltzmann machine more difficult and further limiting the accuracy and efficiency of deep belief network (DBN) model in landslide susceptibility assessment. Therefore, this study proposed a landslide susceptibility assessment method based on Auto-Encoder (AE)-modified DBN. Zhouqu County, Gansu Province in the People’s Republic of China, was selected as the study area. Historical landslides in Zhouqu County were identified using small baseline subset interferometric synthetic aperture radar technology and optical image. Landslide factors were screened based on a geographical detector and stepwise regression method. The Logcosh loss function and determinant coefficient R 2 index were used to evaluate the training process of the AE model, and the balanced cross entropy loss function was used to evaluate the entire network training process. In addition, the area under the curve (AUC) of the synthetical index model (SIM), support vector machine (SVM), and multilayer perceptron (MLP) were compared and evaluated. The results indicated that the proposed model could significantly improve the accuracy of landslide susceptibility assessment. The AUC value of the proposed model was 0.31, 0.12, and 0.11 higher than that of SIM, SVM, and MLP, respectively. Therefore, the improved DBN model based on AE proposed is reliable for early landslide identification and prediction.

RIMD: A novel method for clinical prediction

Electronic health records (EHR) are sparse, noisy, and private, with variable vital measurements and stay lengths. Deep learning models are the current state of the art in many machine learning domain; however, the EHR data is not a suitable training input for most of them. In this paper, we introduce RIMD, a novel deep learning model that consists of a decay mechanism, modular recurrent networks, and a custom loss function that learns minor classes. The decay mechanism learns from patterns in sparse data. The modular network allows multiple recurrent networks to pick only relevant input based on the attention score at a given timestamp. Finally, the custom class balance loss function is responsible for learning minor classes based on samples provided in training. This novel model is used to evaluate predictions for early mortality identification, length of stay, and acute respiratory failure on MIMIC-III dataset. Experiment results indicate that the proposed models outperform similar models in F1-Score, AUROC, and PRAUC scores.

On Some Classes of Estimators Derived from the Positive Part of James–Stein Estimator

This work consists of developing shrinkage estimation strategies for the multivariate normal mean when the covariance matrix is diagonal and known. The domination of the positive part of James–Stein estimator (PPJSE) over James–Stein estimator (JSE) relative to the balanced loss function (BLF) is analytically proved. We introduce a new class of shrinkage estimators which ameliorate the PPJSE, and then we construct a series of polynomial shrinkage estimators which improve the PPJSE; also, any estimator of this series can be ameliorated by adding to it a new term of higher degree. We end this paper by simulation studies which confirm the performance of the suggested estimators.

Synthetic PET via Domain Translation of 3-D MRI.

Historically, patient datasets have been used to develop and validate various reconstruction algorithms for PET/MRI and PET/CT. To enable such algorithm development, without the need for acquiring hundreds of patient exams, in this article we demonstrate a deep learning technique to generate synthetic but realistic whole-body PET sinograms from abundantly available whole-body MRI. Specifically, we use a dataset of 56 18F-FDG-PET/MRI exams to train a 3-D residual UNet to predict physiologic PET uptake from whole-body T1-weighted MRI. In training, we implemented a balanced loss function to generate realistic uptake across a large dynamic range and computed losses along tomographic lines of response to mimic the PET acquisition. The predicted PET images are forward projected to produce synthetic PET (sPET) time-of-flight (ToF) sinograms that can be used with vendor-provided PET reconstruction algorithms, including using CT-based attenuation correction (CTAC) and MR-based attenuation correction (MRAC). The resulting synthetic data recapitulates physiologic 18F-FDG uptake, e.g., high uptake localized to the brain and bladder, as well as uptake in liver, kidneys, heart, and muscle. To simulate abnormalities with high uptake, we also insert synthetic lesions. We demonstrate that this sPET data can be used interchangeably with real PET data for the PET quantification task of comparing CTAC and MRAC methods, achieving ≤ 7.6% error in mean-SUV compared to using real data. These results together show that the proposed sPET data pipeline can be reasonably used for development, evaluation, and validation of PET/MRI reconstruction methods.

AN EXTENDED STUDY TO DETERMINE THE BEST LOSS FUNCTIONS FOR ESTIMATING THE EXPONENTIAL DISTRIBUTION PARAMETER UNDERN EXTENDED STUDY TO DETERMINE THE BEST LOSS FUNCTIONS FOR ESTIMATING THE EXPONENTIAL DISTRIBUTION PARAMETER UNDER JEFFERY AND GAMMA PRIORS

In this research, we compared the Bayesian estimators when estimating the scale parameter for the exponential distribution by using different loss functions under Jeffrey and Gamma priors, as most of the available symmetric and asymmetric loss functions were used, also the balanced and unbalanced loss functions. The simulation results proved the advantage of balanced loss functions with the Gamma prior, and the effectiveness of the balanced loss functions when using Jeffrey prior especially if the value of the weighted coefficient is equal to 0.5, so it is possible to use initial estimators as maximum likelihood estimator to compensate for the lack of prior information around the parameter to be estimated, also the advantage of the balanced general entropy loss function and the balanced weighted square error loss function under Jeffrey prior when the value of the scale parameter for the exponential distribution is less than 1, the preference of the balanced weighted square error loss function and the balanced K loss function if the value of the scale parameter for the exponential distribution is equal to 1, and the preference for the AL-Sayyad balanced loss function and the balanced AL-Bayyati loss function if the value of the scale parameter for the exponential distribution is greater or equal to 2.

BAYESIAN ANALYSIS UNDER UNBALANCED AND BALANCED LOSS FUNCTIONS APPLYING DIFFERENT PRIOR INFORMATIONS

In this paper, We perform a Bayesian analysis of Zeghdoudi distribution based on type II censored data. Using two type of loss functions; balanced and unbalanced loss functions, we use three different loss functions. this estimation includes three cases of prior informations; availability and lack of primary information, we obtain Bayes estimators and the corresponding posterior risks. the analytical forms of these estimators are out of reach, so, we propose Markov chain Monte-Carlo (MCMC) procedure. Moreover, given initial values for the parameters of the model,we obtain maximum likelihood estimators. Furthermore, we compare their performance with those of the Bayesian estimators using balanced and unbalanced loss functions.

Mining Mobile Network Fraudsters with Augmented Graph Neural Networks.

With the rapid evolution of mobile communication networks, the number of subscribers and their communication practices is increasing dramatically worldwide. However, fraudsters are also sniffing out the benefits. Detecting fraudsters from the massive volume of call detail records (CDR) in mobile communication networks has become an important yet challenging topic. Fortunately, Graph neural network (GNN) brings new possibilities for telecom fraud detection. However, the presence of the graph imbalance and GNN oversmoothing problems makes fraudster detection unsatisfactory. To address these problems, we propose a new fraud detector. First, we transform the user features with the help of a multilayer perceptron. Then, a reinforcement learning-based neighbor sampling strategy is designed to balance the number of neighbors of different classes of users. Next, we perform user feature aggregation using GNN. Finally, we innovatively treat the above augmented GNN as weak classifier and integrate multiple weak classifiers using the AdaBoost algorithm. A balanced focal loss function is also used to monitor the model training error. Extensive experiments are conducted on two open real-world telecom fraud datasets, and the results show that the proposed method is significantly effective for the graph imbalance problem and the oversmoothing problem in telecom fraud detection.

Balanced Loss Function for Accurate Surface Defect Segmentation

The accurate image segmentation of surface defects is challenging for modern convolutional neural networks (CNN)-based segmentation models. This paper identifies that loss imbalance is a critical problem in segmentation accuracy improvement. The loss imbalance problem includes: label imbalance, which impairs the accuracy on less represented classes; easy–hard example imbalance, which misleads the focus of optimization on less valuable examples; and boundary imbalance, which involves an unusually large loss value at the defect boundary caused by label confusion. In this paper, a novel balanced loss function is proposed to address the loss imbalance problem. The balanced loss function includes dynamical class weighting, truncated cross-entropy loss and label confusion suppression to solve the three types of loss imbalance, respectively. Extensive experiments are performed on surface defect benchmarks and various CNN segmentation models in comparison with other commonly used loss functions. The balanced loss function outperforms the counterparts and brings accuracy improvement from 5% to 30%.

On E-Bayesian analysis of the hierarchical normal and inverse gamma model using different loss functions and its application

This paper introduces the idea of hierarchical modelling to derive the variance (i.e. squared scale) parameter of the hierarchical normal and inverse gamma model using E-Bayesian estimation. We propose the idea of a hierarchical probability density function instead of the traditional hierarchical prior density function. We aim to derive E-Bayesian estimators with respect to the conjugate prior distribution IG(a,b) on the basis of the Balanced squared error loss function (BSELF) and Stein's loss function (STLF) for the unknown squared scale parameter. We use E-Posterior Risk as an evaluation standard. This study also intends to reveal the relationship among the E-Bayesian estimates under three distinct bivariate independent prior distributions of hyperparameters. The asymptotic properties of the E-Bayesian estimators are also evaluated. Monte Carlo simulations are prosecuted to study the efficiency of E-Bayesian estimators empirically and also, a real data application is analysed for exemplifying purposes.

Analysis of information measures using generalized type-Ⅰ hybrid censored data

<abstract><p>An entropy measure of uncertainty has a complementary dual function called extropy. In the last six years, this measure of randomness has gotten a lot of attention. It cannot, however, be applied to systems that have survived for some time. As a result, the idea of residual extropy was created. To estimate the extropy and residual extropy, Bayesian and non-Bayesian estimators of unknown parameters of the exponentiated gamma distribution are generated. Bayesian estimators are regarded using balanced loss functions like the balanced squared error, balanced linear exponential and balanced general entropy. We use the Lindley method to get the extropy and residual extropy estimates for the exponentiated gamma distribution based on generalized type-Ⅰ hybrid censored data. To test the effectiveness of the proposed methodologies, a simulation experiment was carried out, and the actual data set was studied for illustrative purposes. In summary, the mean squared error values decrease as the number of failures increases, according to the results obtained. The Bayesian estimates of residual extropy under the balanced linear exponential loss function perform well compared to the other estimates. Alternatively, the Bayesian estimates of the extropy perform well under a balanced general entropy loss function in the majority of situations.</p></abstract>