Balanced Loss Research Articles

Bi-directional ConvLSTM residual U-Net retinal vessel segmentation algorithm with improved focal loss function

Accurate blood vessel segmentation on retinal blood vessel images is helpful for the early detection of ophthalmic diseases such as diabetes, hypertension, cardiovascular and cerebrovascular diseases, and inhibits the deterioration of the disease. In current research within the field of retinal blood vessel segmentation, significant challenges exist in accurately segmenting small blood vessels and maintaining blood vessel continuity. The segmentation algorithm proposed in this article offers substantial improvements to address these issues. To enhance the segmentation performance of retinal blood vessels and facilitate more accurate diagnosis of fundus diseases by ophthalmologists, this paper introduces a novel bidirectional convolutional long short-term memory (LSTM) residual U-Net segmentation algorithm, incorporating improvements to the Focal loss function. Firstly, in the encoding part of U-Net, the multi-scale convolution kernels and Bi-ConvLSTM were adopted to improve the residual structure, obtain richer blood vessel features and enhance the detection ability of micro vessels and the continuity of blood vessel characteristics. At the same time, the class balanced cross entropy loss function was improved and the proportional modulation factor is introduced to enhance the learning ability of the network for difficult samples. By adding the Bi-ConvLSTM to the residual structure and introducing the proportional modulation coefficient to the loss function, the network structure realizes better feature information detection and greatly enhances the detection ability of small blood vessels. The experimental analysis on the DRIVE and CHASE_DB1 data sets showed that the sensitivity, specificity, accuracy and AUC reached 0.7961, 0.9796, 0.9563, 0.9792; 0.8344, 0.9665, 0.9547, 0.9758, respectively. The experimental results fully show that the Bi-ConvLSTM residual U-Net segmentation algorithm based on the improved Focal loss function enhances the detection ability of small blood vessel features, improves the continuity of blood vessel features and the network segmentation performance, and is superior to U-Net algorithm and some current mainstream retinal blood vessel segmentation algorithms.

Bayesian inference for the inverse Weibull distribution based on symmetric and asymmetric balanced loss functions with application

In this study, the unified hybrid censored approach is employed to estimate the parameters of the inverse Weibull distribution, as well as the survival and hazard rate functions. Parameter estimates are obtained using both Bayesian and Maximum Likelihood approaches, with Bayesian estimates acquired through Lindley's approximation method using three distinct balanced loss functions. These encompass both symmetric and asymmetric balanced loss functions, specifically the balanced squared error (BSE) loss function, the balanced linear exponential (BLINEX) loss function, and the balanced general entropy (BGE) loss function. We conduct a simulation study to compare the effectiveness of various estimators, and a real-world data analysis is presented to illustrate practical implementation. Ultimately, our findings indicate that Bayesian parameter estimates consistently outperform their Maximum Likelihood counterparts across all methods.

Experimental Demonstration of Controllable PT and Anti-PT Coupling in a Non-Hermitian Metamaterial.

Non-Hermiticity has recently emerged as a rapidly developing field due to its exotic characteristics related to open systems, where the dissipation plays a critical role. In the presence of balanced energy gain and loss with environment, the system exhibits parity-time (PT) symmetry, meanwhile as the conjugate counterpart, anti-PT symmetry can be achieved with dissipative coupling within the system. Here, we demonstrate the coherence of complex dissipative coupling can control the transition between PT and anti-PT symmetry in an electromagnetic metamaterial. Notably, the achievement of the anti-PT symmetric phase is independent of variations in dissipation. Furthermore, we observe phase transitions as the system crosses exceptional points in both anti-PT and PT symmetric metamaterial configurations, achieved by manipulating the frequency and dissipation of resonators. This work provides a promising metamaterial design for broader exploration of non-Hermitian physics and practical application with a controllable Hamiltonian.

In-plane elastic waves in piezoelectric metamaterials with Parity–Time symmetry

The propagation of in-plane waves in piezoelectric metamaterials involves the coupling of longitudinal (i.e. quasi–pressure), transverse (i.e. quasi–shear) and electric potential waves, which can result in different exotic phenomena. In this study, the stiffness matrix method is used to analyze the dispersion relation, which contains anomalous propagation characteristics. With the coupling of electric potential, the frequency spectrum of in–plane wave shows new characteristics. The exceptional point caused by non–Hermitian operator that does not exist before also appears when the vertical wave number is imaginary. The oblique incidence of the in–plane wave at the interface between a homogeneous medium and phononic crystal can excite multiple refraction modes which include the negative refraction. The normal mode decomposition is developed to study the in-plane incidence of piezoelectric metamaterial and then the averaged Poynting vector is applied to obtain the refraction angle. Moreover, the defect layer is introduced into the phononic crystals to study the transmission coefficients with two different symmetrical arrangements. When the balanced gain and loss are considered in the metamaterial, the Parity–Time symmetry appears and brings in pairs of exceptional points to achieve complete transmission with unidirectional zero reflection.

Machine learning-Based Energy Efficient and Enhancing Communication Reliability for MANETs of Balanced Less Loss Routing Protocol

The aim of the presented research was the utilization of machine learning methods, for example, SOM, MARL, and DQN, to promote the increase of energy efficiency and communication reliability in MANET. For the purpose of the research conduction, 10 trials of the experiment were carried out with the view to establish the performance of the proposed approaches from various perspectives. DQN demonstrated superior performance throughout all experiment trials as compared to both SOM and MARL. The model recorded an average energy efficiency of 95%, indicating that it was highly successful in optimizing routing strategies and communication policies. Furthermore, the average packet delivery ratio was shown to be 96%, meaning that DQN provided guarantees of timely and reliable data exchange across the MANET infrastructure. Finally, the average delay was determined to be from 4 ms to 9 ms, evidencing the quick delivery of packets with little temporal latency. The results obtained demonstrate that DQN is able to alleviate the challenges associated with the energy efficiency and communication reliability of MANETs. Notably, deep reinforcement learning with DQN appears to offer viable solutions for more energy-efficient routing, minimized energy consumption in these networks, and improvement to their communication reliability. Thus, the study under consideration has contributed to the development of the existing knowledge as it has expanded the existing understanding of the fact that machine learning, in general, and DQN, in particular, can be utilized to optimize the operation of MANETs.

Sequential estimation for the multiple linear regression models with balanced loss functions

Sequential analysis (SA) as a sampling technique has notable advantages like smaller average sample size and reduced value of risk compared to similarly comparable fixed-sample techniques. In this study, we first propose a few models for the estimation of the regression parameters or functions of parameters under the multiple linear regression (MLR) setup using a balanced loss function (BLF). Thereafter, we obtain the expressions of risk functions and optimal fixed sample sizes for the proposed models based on the bounded risk criteria. We establish that no fixed-sample procedures can tackle these estimation problems. Hence, we propose different multistage sampling methodologies, viz. (i) two-stage sampling, (ii) three-stage sampling, (iii) purely sequential sampling, and (iv) batch sequential sampling, and corroborate the same with the detailed simulation and real data analyses.

Contour wavelet diffusion: A fast and high‐quality image generation model

AbstractDiffusion models can generate high‐quality images and have attracted increasing attention. However, diffusion models adopt a progressive optimization process and often have long training and inference time, which limits their application in realistic scenarios. Recently, some latent space diffusion models have partially accelerated training speed by using parameters in the feature space, but additional network structures still require a large amount of unnecessary computation. Therefore, we propose the Contour Wavelet Diffusion method to accelerate the training and inference speed. First, we introduce the contour wavelet transform to extract anisotropic low‐frequency and high‐frequency components from the input image, and achieve acceleration by processing these down‐sampling components. Meanwhile, due to the good reconstructive properties of wavelet transforms, the quality of generated images can be maintained. Second, we propose a Batch‐normalized stochastic attention module that enables the model to effectively focus on important high‐frequency information, further improving the quality of image generation. Finally, we propose a balanced loss function to further improve the convergence speed of the model. Experimental results on several public datasets show that our method can significantly accelerate the training and inference speed of the diffusion model while ensuring the quality of generated images.

Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data

Learning from open-world noisy data, where both closed-set and open-set noise co-exist in the dataset, is a realistic but underexplored setting. Only recently, several efforts have been initialized to tackle this problem. However, these works assume the classes are balanced when dealing with open-world noisy data. This assumption often violates the nature of real-world large-scale datasets, where the label distributions are generally long-tailed, i.e. class-imbalanced. In this paper, we study the problem of robust visual recognition with class-imbalanced open-world noisy data. We propose a probabilistic graphical model-based approach: iMRF to achieve label noise correction that is robust to class imbalance via an efficient iterative inference of a Markov Random Field (MRF) in each training mini-batch. Furthermore, we design an agreement-based thresholding strategy to adaptively collect clean samples from all classes that includes corrected closed-set noisy samples while rejecting open-set noisy samples. We also introduce a noise-aware balanced cross-entropy loss to explicitly eliminate the bias caused by class-imbalanced data. Extensive experiments on several benchmark datasets including synthetic and real-world noisy datasets demonstrate the superior performance robustness of our method over existing methods. Our code is available at https://github.com/Na-Z/LIOND.

Parity-time imbalance induced by balanced gain and loss medium in non-Hermitian photonic crystals

Photonic crystals, including gain and loss materials, show numerous intriguing features than the ordinary photonic crystals. In this work, using the full wave simulation, we numerically investigate the photonic crystals with gain and loss which are satisfying the parity-time (PT) symmetry condition to reveal the effect of PT symmetry on the optical properties of photonic crystals, including the exceptional point, distribution of electromagnetic fields and edge states. This work reveals a novel mechanism of coexistence and competition between topological states and non-Hermiticity in all-dielectric photonic crystals and provides an innovative understanding of non-Hermiticity in topological photonic systems.

S-NiSe/HG Nanocomposites with Balanced Dielectric Loss Encapsulated in Room-Temperature Self-Healing Polyurethane for Microwave Absorption and Corrosion Protection.

Exploring anticorrosion electromagnetic wave (EMW) absorbing materials in harsh conditions remains a challenge. Herein, S-NiSe/HG nanocomposites encapsulated in room-temperature self-healing polyurethane (S-NiSe/HG/SPU) were exploited as superior anticorrosion EMW absorbing materials. A dual-defect engineering collaborative Schottky interface construction endows S-NiSe/HG with a high vacancy concentration, abundant defects, and moderate conductivity. These structural merits synergistically balance dielectric loss by enhancing dipole-interface polarization loss and optimizing conduction loss. As a result, S-NiSe/HG demonstrates the optimal EMW absorption performance with a minimum reflection loss (RLmin) of -54.8 dB and an adequate absorption bandwidth (EAB) of 7.1 GHz. Besides, S-NiSe/HG/SPU combines the maze effect of S-NiSe/HG with the active repair capability of SPU, thereby providing long-term corrosion resistance for the Mg alloy. Even under corrosion for 10 days, S-NiSe/HG/SPU affords a low corrosion current density (1.3 × 10-5 A) and high charge transfer resistance (3796 Ω cm2). Overall, this work provides valuable insights for in-depth exploration of dielectric loss and development of multifunctional EMW-absorbing materials.

Low and high dimensional wavelet thresholds for matrix-variate normal distribution

The matrix-variate normal distribution is a probability distribution that is a generalization of the multivariate normal distribution to matrix-valued random variables. In this paper, we introduce a wavelet shrinkage estimator based on Stein’s unbiased risk estimate (SURE) threshold for matrix-variate normal distribution. We find a new SURE threshold for soft thresholding wavelet shrinkage estimator under the reflected normal balanced loss function in low and high dimensional cases. Also, we obtain the restricted wavelet shrinkage estimator based on non-negative sub matrix of the mean matrix. Finally, we present a simulation study to test the validity of the wavelet shrinkage estimator and two real examples for low and high dimensional data sets.

A Comprehensive Framework for Long-Tailed Learning via Pretraining and Normalization.

Data in the visual world often present long-tailed distributions. However, learning high-quality representations and classifiers for imbalanced data is still challenging for data-driven deep learning models. In this work, we aim at improving the feature extractor and classifier for long-tailed recognition via contrastive pretraining and feature normalization, respectively. First, we carefully study the influence of contrastive pretraining under different conditions, showing that current self-supervised pretraining for long-tailed learning is still suboptimal in both performance and speed. We thus propose a new balanced contrastive loss and a fast contrastive initialization scheme to improve previous long-tailed pretraining. Second, based on the motivative analysis on the normalization for classifier, we propose a novel generalized normalization classifier that consists of generalized normalization and grouped learnable scaling. It outperforms traditional inner product classifier as well as cosine classifier. Both the two components proposed can improve recognition ability on tail classes without the expense of head classes. We finally build a unified framework that achieves competitive performance compared with state of the arts on several long-tailed recognition benchmarks and maintains high efficiency.

Meal Frequency Does Not Affect Weight Loss in Overweight/Obese Women but Affects the Body Composition: A Randomized Controlled Trial

Objectives Studies showing the relationship between meal frequency, weight loss and anthropometric measurements are contradictory. This study aims to determine the effect of meal frequency on weight loss, anthropometric measurements, and body composition. Methods This is a parallel designed randomized control trial that was conducted with 40 female volunteers between the ages of 19–64 years, with a Body Mass Index (BMI) ≥27 who applied to a private clinic. Participants were randomized in two treatment arms (3 meals + 3 snacks/day (n = 20) vs 3 meals/day (n = 20)) and same dietary energy restriction (–500kcal) was applied for 3 months. Food consumption was questioned with 3 day food records, and anthropometric measurements and body composition were measured before the study and repeated each week till the end of the study by the researcher. Results All of the participants completed the study period. Body weight (kg), BMI (kg/m2), total body fat (kg), body fat percentage (%), fat free mass (kg) and waist circumference (cm) decreased, while fat free mass percentage (%) increased significantly in both of the groups at the end of the study (p < 0.05). The rate of difference for body weight, BMI (kg/m2) and waist circumference (cm) were similar among the groups. When difference in body composition analyses was examined, the rate of reduction in total body fat (–18.82 ± 4.97% vs −14.87 ± 7.44%) and body fat percentage (%)(–10.79 ± 4.63% vs −7.68 ± 7.04%) and the rate of increase in fat free mass percentage (%)(7.65 ± 3.16% vs 5.04 ± 3.44%) were significantly higher in 3 meals + 3 snacks group (p < 0.05). Conclusion When energy restricted and balanced weight loss programs are applied, alteration in body weight, BMI and waist circumference is not affected from meal frequency, but body composition does. Clinical Trial Number NCT05581862 (Date of Trial Registration: 13/10/2022)

Semantic Segmentation of Surface Cracks in Urban Comprehensive Pipe Galleries Based on Global Attention

Cracks inside urban underground comprehensive pipe galleries are small and their characteristics are not obvious. Due to low lighting and large shadow areas, the differentiation between the cracks and background in an image is low. Most current semantic segmentation methods focus on overall segmentation and have a large perceptual range. However, for urban underground comprehensive pipe gallery crack segmentation tasks, it is difficult to pay attention to the detailed features of local edges to obtain accurate segmentation results. A Global Attention Segmentation Network (GA-SegNet) is proposed in this paper. The GA-SegNet is designed to perform semantic segmentation by incorporating global attention mechanisms. In order to perform precise pixel classification in the image, a residual separable convolution attention model is employed in an encoder to extract features at multiple scales. A global attention upsample model (GAM) is utilized in a decoder to enhance the connection between shallow-level features and deep abstract features, which could increase the attention of the network towards small cracks. By employing a balanced loss function, the contribution of crack pixels is increased while reducing the focus on background pixels in the overall loss. This approach aims to improve the segmentation accuracy of cracks. The comparative experimental results with other classic models show that the GA SegNet model proposed in this study has better segmentation performance and multiple evaluation indicators, and has advantages in segmentation accuracy and efficiency.

CoDC: Accurate Learning with Noisy Labels via Disagreement and Consistency.

Inspired by the biological nervous system, deep neural networks (DNNs) are able to achieve remarkable performance in various tasks. However, they struggle to handle label noise, which can poison the memorization effects of DNNs. Co-teaching-based methods are popular in learning with noisy labels. These methods cross-train two DNNs based on the small-loss criterion and employ a strategy using either "disagreement" or "consistency" to obtain the divergence of the two networks. However, these methods are sample-inefficient for generalization in noisy scenarios. In this paper, we propose CoDC, a novel Co-teaching-basedmethod for accurate learning with label noise via both Disagreement and Consistency strategies. Specifically, CoDC maintains disagreement at the feature level and consistency at the prediction level using a balanced loss function. Additionally, a weighted cross-entropy loss is proposed based on information derived from the historical training process. Moreover, the valuable knowledge involved in "large-loss" samples is further developed and utilized by assigning pseudo-labels. Comprehensive experiments were conducted on both synthetic and real-world noise and under various noise types. CoDC achieved 72.81% accuracy on the Clothing1M dataset and 76.96% (Top1) accuracy on the WebVision1.0 dataset. These superior results demonstrate the effectiveness and robustness of learning with noisy labels.

BWLM: A Balanced Weight Learning Mechanism for Long-Tailed Image Recognition

With the growth of data in the real world, datasets often encounter the problem of long-tailed distribution of class sample sizes. In long-tailed image recognition, existing solutions usually adopt a class rebalancing strategy, such as reweighting based on the effective sample size of each class, which leans towards common classes in terms of higher accuracy. However, increasing the accuracy of rare classes while maintaining the accuracy of common classes is the key to solving the problem of long-tailed image recognition. This research explores a direction that balances the accuracy of both common and rare classes simultaneously. Firstly, a two-stage training is adopted, motivated by the use of transfer learning to balance features of common and rare classes. Secondly, a balanced weight function called Balanced Focal Softmax (BFS) loss is proposed, which combines balanced softmax loss focusing on common classes with balanced focal loss focusing on rare classes to achieve dual balance in long-tailed image recognition. Subsequently, a Balanced Weight Learning Mechanism (BWLM) to further utilize the feature of weight decay is proposed, where the weight decay as the weight balancing technique for the BFS loss tends to make the model learn smaller balanced weights by punishing the larger weights. Through extensive experiments on five long-tailed image datasets, it proves that transferring the weights from the first stage to the second stage can alleviate the bias of the naive models toward common classes. The proposed BWLM not only balances the weights of common and rare classes, but also greatly improves the accuracy of long-tailed image recognition and outperforms many state-of-the-art algorithms.

P T-symmetry breaking in dual-core phosphate-glass photonic crystal fibers.

We investigate the properties of a soft glass dual-core photonic crystal fiber for application in multicore waveguiding with balanced gain and loss. Its base material is a phosphate glass in a P2O5-Al2O3-Yb2O3-BaO-ZnO-MgO-Na2O oxide system. The separated gain and loss cores are realized with two cores with ytterbium and copper doping of the base phosphate glass. The ytterbium-doped core supports a laser (gain) activity under excitation with a pump at 1000 nm wavelength, while the CuO-doped is responsible for strong attenuation at the same wavelength. We establish conditions for an exact balance between gain and loss and investigate pulse propagation by solving a system of coupled generalized nonlinear Schrödinger equations. We predict two states of light under excitation with hyperbolic secant pulses centered at 1000 nm: 1) linear oscillation of the pulse energy between gain and loss core (P T-symmetry state), with strong power attenuation; 2) retention of the pulse in the excited gain core (broken P T-symmetry), with very modest attenuation. The optimal pulse energy levels were identified to be 100 pJ (first state) and 430 pJ (second state).

Bayesian estimation of the Pareto model based on type-II censoring data by employing non-linear programming

The main goal of this article is to determine the optimally weighted coefficients (Ω1and Ω2) of the balanced loss function of the form. LΚ,Ω,ξoΨ(σ),ξ=Ω1γσΚξo,ξ+Ω2γσΚΨ(σ),ξ;Ω1+Ω2=1. Based on Type II Censored Data, by applying non-linear programming to estimate the shape parameter and some survival time characteristics, such as reliability and hazard functions of the Pareto distribution. Considering two balanced loss functions (BLF), including balanced square error loss function (BSELF) and balanced linear exponential loss function (BLLF), the calculation is based on the balanced loss function, including symmetric and asymmetric loss functions, as a special case. Use Monte Carlo simulation to pass Bayesian and maximum likelihood estimators through. The results of the simulation showed that the proposed model BLLF has the best performance. Moreover, the simulation verified that the balanced loss functions are always better than the corresponding loss function.

Bayes Estimation for a Lifetime Distribution of the Next Class of Balanced Loss

Bayes Estimation for a Lifetime Distribution of the Next Class of Balanced Loss

Inference of stress-strength reliability based on adaptive progressive type-Ⅱ censing from Chen distribution with application to carbon fiber data

&lt;abstract&gt;&lt;p&gt;In this paper, we used the maximum likelihood estimation (MLE) and the Bayes methods to perform estimation procedures for the reliability of stress-strength $ R = P(Y &amp;lt; X) $ based on independent adaptive progressive censored samples that were taken from the Chen distribution. An approximate confidence interval of $ R $ was constructed using a variety of classical techniques, such as the normal approximation of the MLE, the normal approximation of the log-transformed MLE, and the percentile bootstrap (Boot-p) procedure. Additionally, the asymptotic distribution theory and delta approach were used to generate the approximate confidence interval. Further, the Bayesian estimation of $ R $ was obtained based on the balanced loss function, which came in two versions here, the symmetric balanced squared error (BSE) loss function and the asymmetric balanced linear exponential (BLINEX) loss function. When estimating $ R $ using the Bayesian approach, all the unknown parameters of the Chen distribution were assumed to be independently distributed and to have informative gamma priors. Additionally, a mixture of Gibbs sampling algorithm and Metropolis-Hastings algorithm was used to compute the Bayes estimate of $ R $ and the associated highest posterior density credible interval. In the end, simulation research was used to assess the general overall performance of the proposed estimators and a real dataset was provided to exemplify the theoretical results.&lt;/p&gt;&lt;/abstract&gt;