Specific Loss Function Research Articles

Differential evolution-based neural architecture search for brain vessel segmentation

Brain vasculature analysis is critical in developing novel treatment targets for neurodegenerative diseases. Such an accurate analysis cannot be performed manually but requires a semi-automated or fully-automated approach. Deep learning methods have recently proven indispensable for the automated segmentation and analysis of medical images. However, optimizing a deep learning network architecture is another challenge. Manually selecting deep learning network architectures and tuning their hyper-parameters requires a lot of expertise and effort. To solve this problem, neural architecture search (NAS) approaches that explore more efficient network architectures with high segmentation performance have been proposed in the literature. This study introduces differential evolution-based NAS approaches in which a novel search space is proposed for brain vessel segmentation. We select two architectures that are frequently used for medical image segmentation, i.e. U-Net and Attention U-Net, as baselines for NAS optimizations. The conventional differential evolution and the opposition-based differential evolution with novel search space are employed as search methods in NAS. Furthermore, we perform ablation studies and evaluate the effects of specific loss functions, model pruning, threshold selection and generalization performance on the proposed models. The experiments are conducted on two datasets providing 335 single-channel 8-bit gray-scale images. These datasets are a public volumetric cerebrovascular system dataset (vesseINN) and our own dataset called KUVESG. The proposed NAS approaches, namely UNAS-Net and Attention UNAS-Net architectures, yield better segmentation performance in terms of different segmentation metrics. More specifically, UNAS-Net with differential evolution reveals high dice score/sensitivity values of 79.57/81.48, respectively. Moreover, they provide shorter inference times by a factor of 9.15 than the baseline methods.

An Optimization-Based Order-and-Cut Approach for Fair Clustering of Data Sets

Machine learning algorithms have been increasingly integrated into applications that significantly affect human lives. This surged an interest in designing algorithms that train machine learning models to minimize training error and imposing a certain level of fairness. In this paper, we consider the problem of fair clustering of data sets. In particular, given a set of items each associated with a vector of nonsensitive attribute values and a categorical sensitive attribute (e.g., gender, race, etc.), our goal is to find a clustering of the items that minimizes the loss (i.e., clustering objective) function and imposes fairness measured by Rényi correlation. We propose an efficient and scalable in-processing algorithm, driven by findings from the field of combinatorial optimization, that heuristically solves the underlying optimization problem and allows for regulating the trade-off between clustering quality and fairness. The approach does not restrict the analysis to a specific loss function, but instead considers a more general form that satisfies certain desirable properties. This broadens the scope of the algorithm’s applicability. We demonstrate the effectiveness of the algorithm for the specific case of k-means clustering as it is one of the most extensively studied and widely adopted clustering schemes. Our numerical experiments reveal the proposed algorithm significantly outperforms existing methods by providing a more effective mechanism to regulate the trade-off between loss and fairness. History: Rema Padman served as the senior editor for this article. Data Ethics & Reproducibility Note: The code capsule is available on Code Ocean at https://codeocean.com/capsule/9556728/tree and in the e-Companion to the this article (available at https://doi.org/10.1287/ijds.2022.0005 ).

Multi-Resolution Edge-aware Lighting Enhancement Network

Images taken under poor lighting conditions generally present problems such as low brightness, noise, and color distortion, so dealing with low-light images is challenging. To solve this problem, current mainstream methods focus on designing special deep networks. For example, many methods are based on Retinex-based models. However, the solution of the Retinex model is an ill-conditioned problem in mathematics, so these methods are difficult to obtain the expected result. In this paper, following the new paradigm of embedding prior knowledge in deep learning, we design a simple yet surprisingly effective neural network, called Multi-Resolution Edge-aware Lighting Enhancement Network (MELENet). Since resolution and acutance are two of the key factors related to the visual quality of natural images, MELENet with a dual-branch structure aims to incorporate multi-resolution edge information into deep learning for guiding lighting enhancement. Additionally, by adopting special loss functions, the model suppresses noise and prevents color distortion while obtaining sound lighting enhancement effects. Experiments have verified that such a simple solution exhibits outstanding performance compared to existing mainstream lighting enhancement methods. We also directly apply the proposed model to image de-hazing and obtain state-of-the-art results.

GIPC-GAN: an end-to-end gradient and intensity joint proportional constraint generative adversarial network for multi-focus image fusion

As for the problems of boundary blurring and information loss in the multi-focus image fusion method based on the generative decision maps, this paper proposes a new gradient-intensity joint proportional constraint generative adversarial network for multi-focus image fusion, with the name of GIPC-GAN. First, a set of labeled multi-focus image datasets using the deep region competition algorithm on a public dataset is constructed. It can train the network and generate fused images in an end-to-end manner, while avoiding boundary errors caused by artificially constructed decision maps. Second, the most meaningful information in the multi-focus image fusion task is defined as the target intensity and detail gradient, and a jointly constrained loss function based on intensity and gradient proportional maintenance is proposed. Constrained by a specific loss function to force the generated image to retain the information of target intensity, global texture and local texture of the source image as much as possible and maintain the structural consistency between the fused image and the source image. Third, we introduce GAN into the network, and establish an adversarial game between the generator and the discriminator, so that the intensity structure and texture gradient retained by the fused image are kept in a balance, and the detailed information of the fused image is further enhanced. Last but not least, experiments are conducted on two multi-focus public datasets and a multi-source multi-focus image sequence dataset and compared with other 7 state-of-the-art algorithms. The experimental results show that the images fused by the GIPC-GAN model are superior to other comparison algorithms in both subjective performance and objective measurement, and basically meet the requirements of real-time image fusion in terms of running efficiency and mode parameters quantity.

Systematic Review of Deep Learning Approaches for Automatic Segmentation of Abdominal Aortic Aneurysm and Thrombus on Computed Tomography Angiography Images

Abdominal Aortic Aneurysm (AAA) is a potentially life-threatening condition characterized by the enlargement of the abdominal aorta. Computed Tomography Angiography (CTA) is a widely used diagnostic tool for AAA, and accurate segmentation of the aneurysm and thrombus is critical for treatment planning. Deep learning approaches have shown promise in automating the segmentation process. We conducted a systematic review of the literature to evaluate the performance of deep learning methods for automatic segmentation of AAA and thrombus on CTA images. Six studies were identified that met our inclusion criteria. The studies utilized various deep learning architectures and loss functions to segment AAA and thrombus, and reported performance using metrics such as sensitivity, specificity, accuracy, and Dice coefficient. The results indicate that deep learning methods can achieve high accuracy and Dice coefficient values for segmentation of AAA and thrombus on CTA images. However, the performance of the methods varied depending on the specific architecture and loss function used. Further research is needed to determine the most effective deep learning approach for automatic segmentation of AAA and thrombus on CTA images.

Infrared and visible image fusion based on a two-stage class conditioned auto-encoder network

Existing auto-encoder based infrared and visible image fusion methods typically utilize a shared encoder to extract features from different modalities and adopt a handcrafted fusion strategy to fuse the extracted features into intermediate representation before the decoder part. In this paper, we present a novel two-stage class conditioned auto-encoder framework for high-quality multispectral fusion tasks. In the first training stage, we introduce a class embedding sub-branch to the encoder network for modeling the characteristics of different modalities and adaptively scaling the intermediate features based on the input modality. Moreover, we design a cross-transfer residual block to promote the content and texture information flow in the encoder for generating more representative features. In the second training stage, we insert a learnable fusion module between the pre-trained class conditioned encoder and decoder parts to replace the handcrafted fusion strategy. Specific intensity and gradient loss functions are utilized to tune the model for the fusion of distinctive deep features in a data-driven manner. With the important designs including the class conditioned auto-encoder and the two-stage training strategy, our proposed TS-ClassFuse can better preserve distinctive information/features from the source images and decrease the training difficulty for simultaneously extracting informative features and determining the optimal fusion scheme. Experimental results verify the effectiveness of our method in terms of both qualitative and quantitative evaluations.

Efficient virtual high-content screening using a distance-aware transformer model

Molecular similarity search is an often-used method in drug discovery, especially in virtual screening studies. While simple one- or two-dimensional similarity metrics can be applied to search databases containing billions of molecules in a reasonable amount of time, this is not the case for complex three-dimensional methods. In this work, we trained a transformer model to autoencode tokenized SMILES strings using a custom loss function developed to conserve similarities in latent space. This allows the direct sampling of molecules in the generated latent space based on their Euclidian distance. Reducing the similarity between molecules to their Euclidian distance in latent space allows the model to perform independent of the similarity metric it was trained on. While we test the method here using 2D similarity as proof-of-concept study, the algorithm will enable also high-content screening with time-consuming 3D similarity metrics. We show that the presence of a specific loss function for similarity conservation greatly improved the model’s ability to predict highly similar molecules. When applying the model to a database containing 1.5 billion molecules, our model managed to reduce the relevant search space by 5 orders of magnitude. We also show that our model was able to generalize adequately when trained on a relatively small dataset of representative structures. The herein presented method thereby provides new means of substantially reducing the relevant search space in virtual screening approaches, thus highly increasing their throughput. Additionally, the distance awareness of the model causes the efficiency of this method to be independent of the underlying similarity metric.

Fine-Grained Classification of Solder Joints With α-Skew Jensen–Shannon Divergence

Solder joint inspection (SJI) is a critical process in the production of printed circuit boards (PCB). Detection of solder errors during SJI is quite challenging as the solder joints have very small sizes and can take various shapes. In this study, we first show that solders have low feature diversity, and that the SJI can be carried out as a fine-grained image classification task which focuses on hard-to-distinguish object classes. To improve the fine-grained classification accuracy, penalizing confident model predictions by maximizing entropy was found useful in the literature. Inline with this information, we propose using the α-skew Jensen-Shannon divergence (α-JS) for penalizing the confidence in model predictions. We compare the α-JS regularization with both existing entropy-regularization based methods and the methods based on attention mechanism, segmentation techniques, transformer models, and specific loss functions for fine-grained image classification tasks. We show that the proposed approach achieves the highest F1-score and competitive accuracy for different models in the fine-grained solder joint classification task. Finally, we visualize the activation maps and show that with entropy-regularization, more precise class-discriminative regions are localized, which are also more resilient to noise. The code is available at https://github.com/furkanulger/SJI-entropy-reg.

Combining band-frequency separation and deep neural networks for optoacoustic imaging

In this paper we consider the problem of image reconstruction in optoacoustic tomography. In particular, we devise a deep neural architecture that can explicitly take into account the band-frequency information contained in the sinogram. This is accomplished by two means. First, we jointly use a linear filtered back-projection method and a fully dense UNet for the generation of the images corresponding to each one of the frequency bands considered in the separation. Secondly, in order to train the model, we introduce a special loss function consisting of three terms: (i) a separating frequency bands term; (ii) a sinogram-based consistency term and (iii) a term that directly measures the quality of image reconstruction and which takes advantage of the presence of ground-truth images present in training dataset. Numerical experiments show that the proposed model, which can be easily trainable by standard optimization methods, presents an excellent generalization performance quantified by a number of metrics commonly used in practice. Also, in the testing phase, our solution has a comparable (in some cases lower) computational complexity, which is a desirable feature for real-time implementation of optoacoustic imaging.

Privacy-Aware Adversarial Network in Human Mobility Prediction

As mobile devices and location-based services are increasingly developed in different smart city scenarios and applications, many unexpected privacy leakages have arisen due to geolocated data collection and sharing. User re-identification and other sensitive inferences are major privacy threats when geolocated data are shared with cloud-assisted applications. Significantly, four spatio-temporal points are enough to uniquely identify 95% of the individuals, which exacerbates personal information leakages. To tackle malicious purposes such as user re-identification, we propose an LSTM-based adversarial mechanism with representation learning to attain a privacy-preserving feature representation of the original geolocated data (i.e., mobility data) for a sharing purpose. These representations aim to maximally reduce the chance of user re-identification and full data reconstruction with a minimal utility budget (i.e., loss). We train the mechanism by quantifying privacy-utility trade-off of mobility datasets in terms of trajectory reconstruction risk, user re-identification risk, and mobility predictability. We report an exploratory analysis that enables the user to assess this trade-off with a specific loss function and its weight parameters. The extensive comparison results on four representative mobility datasets demonstrate the superiority of our proposed architecture in mobility privacy protection and the efficiency of the proposed privacy-preserving features extractor. We show that the privacy of mobility traces attains decent protection at the cost of marginal mobility utility. Our results also show that by exploring the Pareto optimal setting, we can simultaneously increase both privacy (45%) and utility (32%).

Unsupervised SAR Image Change Detection Based on Feature Fusion of Information Transfer

Synthetic aperture radar (SAR) image change detection is a hot but challenging task due to SAR images’ complex contents and inherent speckle noises. The expected change detection methods should reduce the influence of speckle noises, obtain the discriminative feature representations, and generate accurate change maps simultaneously. To these ends, we propose a new SAR image change detection method named feature fusion of information transfer network (FFITN). First, we develop a hybrid convolution block to depress the speckle noise impacts and explore the valuable information from SAR images. Thus, the feature extraction module (FEM) is constructed to obtain the multi-level features. Then, an information transfer module (ITM) is proposed to capture the salient regions from various aspects. Also, the salient knowledge is transferred among features at different levels to enhance their discrimination. Next, a self-attention-based feature fusion module (SAFFM) is introduced to fuse various features. Finally, a change map generation module (CMGM) with the clustering algorithm and specific loss functions is designed to produce the pseudo labels and change maps. Experimental results on three public SAR data sets demonstrate the model’s effectiveness. Our source codes are available at https://github.com/TangXu-Group/FFITN.

Towards Robust Visual Object Tracking with Independent Target-Agnostic Detection and Effective Siamese Cross-Task Interaction.

Advanced Siamese visual object tracking architectures are jointly trained using pair-wise input images to perform target classification and bounding box regression. They have achieved promising results in recent benchmarks and competitions. However, the existing methods suffer from two limitations: First, though the Siamese structure can estimate the target state in an instance frame, provided the target appearance does not deviate too much from the template, the detection of the target in an image cannot be guaranteed in the presence of severe appearance variations. Second, despite the classification and regression tasks sharing the same output from the backbone network, their specific modules and loss functions are invariably designed independently, without promoting any interaction. Yet, in a general tracking task, the centre classification and bounding box regression tasks are collaboratively working to estimate the final target location. To address the above issues, it is essential to perform target-agnostic detection so as to promote cross-task interactions in a Siamese-based tracking framework. In this work, we endow a novel network with a target-agnostic object detection module to complement the direct target inference, and to avoid or minimise the misalignment of the key cues of potential template-instance matches. To unify the multi-task learning formulation, we develop a cross-task interaction module to ensure consistent supervision of the classification and regression branches, improving the synergy of different branches. To eliminate potential inconsistencies that may arise within a multi-task architecture, we assign adaptive labels, rather than fixed hard labels, to supervise the network training more effectively. The experimental results obtained on several benchmarks, i.e., OTB100, UAV123, VOT2018, VOT2019, and LaSOT, demonstrate the effectiveness of the advanced target detection module, as well as the cross-task interaction, exhibiting superior tracking performance as compared with the state-of-the-art tracking methods.

Design and Analysis of Convolutional Neural Layers: A Signal Processing Perspective

Convolutional layers (CLs) are ubiquitous in contemporary deep neural network (DNN) models, commonly used for automatic feature extraction. A CL performs cross-correlation between the input to the layer and a set of learnable kernels to produce the layer output. Typically, kernel weights are randomly initialized and automatically learned during model training using the backpropagation and gradient descent algorithms to minimize a specific loss function. Modern DNN models comprise deep hierarchical stacks of CLs and pooling layers. Despite their prevalence, CLs are perceived as a magical tool for feature extraction without solid interpretations of their underlying working principles. In this work, we advance a method for designing and analyzing CLs by providing novel signal processing interpretations of the CL by exploiting the correlation and equivalent convolution functions of the layer. The proposed interpretations enable the employment of CLs to develop finite impulse response (FIR) filters, matched filters (MFs), short-time Fourier transform (STFT), discrete-time Fourier transform (DTFT), and continuous wavelet transform (CWT) algorithms. The main idea is to pre-assign the CL kernel weights to implement a specific convolution- or correlation-based DSP algorithm. Such an approach enables building self-contained DNN models in which CLs are utilized for various preprocessing and feature extractions tasks, enhancing the model portability, and cutting down the preprocessing computational cost. The proposed DSP interpretations provide an effective means to analyze and explain the operation of automatically trained CLs in the time and frequency domains by reversing the design procedures. The presented interpretations are mathematically established and experimentally validated with a comprehensive machinery fault diagnosis application example illustrating the potential of the proposed methodology.

Reliable and Intelligent Fault Diagnosis With Evidential VGG Neural Networks

With the emergence of Internet-of-Things and big data technologies, data-driven fault diagnosis approaches, notably deep learning-based methods have shown promising capabilities in achieving high accuracy through end-to-end learning. However, these deterministic neural networks cannot incorporate the prediction uncertainty, which is critical in practical applications with possible out-of-distribution data. The present paper develops a reliable and intelligent fault diagnosis framework based on evidence theory and improved VGG neural networks, which can achieve accurate and reliable diagnosis results by incorporating additional estimation of the prediction uncertainty. Specifically, the paper treats the predictions of the VGG as subjective opinions by placing a Dirichlet distribution on the category probabilities and collecting the evidence from data during the training process. A specific loss function assisted by evidence theory is adopted for the VGG to obtain improved uncertainty estimations. The proposed method, which incorporates Evidential VGG (EVGG) neural networks, as termed here, is verified by a case study of the fault diagnosis of rolling bearings, in the presence of sensing noise and sensor failure. The experimental results illustrate that the proposed method can estimate the prediction uncertainty and avoid overconfidence in fault diagnosis with out-of-distribution (OOD) data. Also, the developed approach is shown to perform robustly under various levels of noise, which indicates a high potential for use in practical applications.

Quantitative Prediction of SYNTAX Score for Cardiovascular Artery Disease Patients via the Inverse Problem Algorithm Technique as Artificial Intelligence Assessment in Diagnostics.

The quantitative prediction of the SYNTAX score for cardiovascular artery disease patients using the inverse problem algorithm (IPA) technique in artificial intelligence was explored in this study. A 29-term semi-empirical formula was defined according to seven risk factors: (1) age, (2) mean arterial pressure, (3) body surface area, (4) pre-prandial blood glucose, (5) low-density-lipoprotein cholesterol, (6) Troponin I, and (7) C-reactive protein. Then, the formula was computed via the STATISTICA 7.0 program to obtain a compromised solution for a 405-patient dataset with a specific loss function [actual-predicted]2 as low as 3.177, whereas 0.0 implies a 100% match between the prediction and observation via "the lower, the better" principle. The IPA technique first created a data matrix [405 × 29] from the included patients' data and then attempted to derive a compromised solution of the column matrix of 29-term coefficients [29 × 1]. The correlation coefficient, r2, of the regression line for the actual versus predicted SYNTAX score was 0.8958, showing a high coincidence among the dataset. The follow-up verification based on another 105 patients' data from the same group also had a high correlation coefficient of r2 = 0.8304. Nevertheless, the verified group's low derived average AT (agreement) (ATavg = 0.308 ± 0.193) also revealed a slight deviation between the theoretical prediction from the STATISTICA 7.0 program and the grades assigned by clinical cardiologists or interventionists. The predicted SYNTAX scores were compared with earlier reported findings based on a single-factor statistical analysis or scanned images obtained by sonography or cardiac catheterization. Cardiologists can obtain the SYNTAX score from the semi-empirical formula for an instant referral before performing a cardiac examination.

PET and MRI image fusion based on a dense convolutional network with dual attention

The fusion techniques of different modalities in medical images, e.g., Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI), are increasingly significant in many clinical applications by integrating the complementary information from different medical images. In this paper, we propose a novel fusion model based on a dense convolutional network with dual attention (CSpA-DN) for PET and MRI images. In our framework, an encoder composed of the densely connected neural network is constructed to extract features from source images, and a decoder network is employed to generate the fused image from these features. Simultaneously, a dual-attention module is introduced in the encoder and decoder to further integrate local features along with their global dependencies adaptively. In the dual-attention module, a spatial attention block is leveraged to extract features of each point from encoder network by a weighted sum of feature information at all positions. Meanwhile, the interdependent correlation of all image features is aggregated via a module of channel attention. In addition, we design a specific loss function including image loss, structural loss, gradient loss and perception loss to preserve more structural and detail information and sharpen the edges of targets. Our approach facilitates the fused images to not only preserve abundant functional information from PET images but also retain rich detail structures of MRI images. Experimental results on publicly available datasets illustrate the superiorities of CSpA-DN model compared with state-of-the-art methods according to both qualitative observation and objective assessment.

Zero-Shot Remote Sensing Image Dehazing Based on a Re-Degradation Haze Imaging Model

Image dehazing is crucial for improving the advanced applications on remote sensing (RS) images. However, collecting paired RS images to train the deep neural networks (DNNs) is scarcely available, and the synthetic datasets may suffer from domain-shift issues. In this paper, we propose a zero-shot RS image dehazing method based on a re-degradation haze imaging model, which directly restores the haze-free image from a single hazy image. Based on layer disentanglement, we design a dehazing framework consisting of three joint sub-modules to disentangle the hazy input image into three components: the atmospheric light, the transmission map, and the recovered haze-free image. We then generate a re-degraded hazy image by mixing up the hazy input image and the recovered haze-free image. By the proposed re-degradation haze imaging model, we theoretically demonstrate that the hazy input and the re-degraded hazy image follow a similar haze imaging model. This finding helps us to train the dehazing network in a zero-shot manner. The dehazing network is optimized to generate outputs that satisfy the relationship between the hazy input image and the re-degraded hazy image in the re-degradation haze imaging model. Therefore, given a hazy RS image, the dehazing network directly infers the haze-free image by minimizing a specific loss function. Using uniform hazy datasets, non-uniform hazy datasets, and real-world hazy images, we conducted comprehensive experiments to show that our method outperforms many state-of-the-art (SOTA) methods in processing uniform or slight/moderate non-uniform RS hazy images. In addition, evaluation on a high-level vision task (RS image road extraction) further demonstrates the effectiveness and promising performance of the proposed zero-shot dehazing method.

Restricted Stein-rule estimation in ultrastructural linear measurement error models

We consider a multivariate ultrastructural measurement error model when some information on regression coefficients is available in the form of linear restriction. We propose Stein-rule estimation for the regression coefficients. Without any assumption about the distribution for the random components, the asymptotic risk under a specific quadratic loss function is studied and a sufficient condition for the dominance of the proposed estimator over a consistent estimator. We also carry out a simulation study to demonstrate the finite sample properties of the estimators.

A Primary-Auxiliary Coupled Neural Network for Three-Dimensional Holographic Particle Field Characterization

Particle field measurement is an important topic in many industrial branches. However, there are always complex imaging scenes in the engineering experiments, resulting in severe imaging artifact, noise, and blur, such as the optical holography. In this article, we propose a primary-auxiliary coupled neural network (PANet) for 3-D holographic particle field characterization, which can obtain a comprehensive particle measurement, including the identification, focus determination, segmentation, and size estimation. PANet is constituted by two subnets that are arranged in a coupled architecture, i.e., a Primary-Net (PNet) and an AuxiliaryNet (ANet). As the main frame, PNet is designed to accomplish the detection of most particles, while ANet aims to detect the tiny particles that PNet cannot identify. We exploit an alternative training method to realize their functional differentiation and complementation. PANet is evaluated on two kinds of holographic particle field data, i.e., high-energy laser shock aluminum target and droplet breakup in high Mach shock wave. By means of semisupervised learning and a specific loss function, the effect of deficient particle labeling can be alleviated. Experimental results demonstrate that PANet can achieve excellent performance in particle field characterization, especially for those with a wide size span and complex image background.

A novel interpretable machine learning algorithm to identify optimal parameter space for cancer growth.

Recent years have seen an increase in the application of machine learning to the analysis of physical and biological systems, including cancer progression. A fundamental downside to these tools is that their complexity and nonlinearity makes it almost impossible to establish a deterministic, a priori relationship between their input and output, and thus their predictions are not wholly accountable. We begin with a series of proofs establishing that this holds even for the simplest possible model of a neural network; the effects of specific loss functions are explored more fully in Appendices. We return to first principles and consider how to construct a physics-inspired model of tumor growth without resorting to stochastic gradient descent or artificial nonlinearities. We derive an algorithm which explores the space of possible parameters in a model of tumor growth and identifies candidate equations much faster than a simulated annealing approach. We test this algorithm on synthetic tumor-growth trajectories and show that it can efficiently and reliably narrow down the area of parameter space where the correct values are located. This approach has the potential to greatly improve the speed and reliability with which patient-specific models of cancer growth can be identified in a clinical setting.