Terms Of Loss Research Articles

A physics-aware neural network for effective refractive index prediction of photonic waveguides

Neural network (NN)—based surrogates have been effectively used for modeling dynamic systems, including photonic devices. However, black-box data-driven modeling approaches significantly suffer from performance reduction in high-dimensional spaces.As a remedy, we propose a novel physics-aware NN architecture for the effective index prediction of photonic strip waveguides. The model learns a translation between the strip waveguide and an equivalent infinite slab waveguide by employing physical loss terms in the loss function. The proposed method exhibits significantly lower error, with more than 50% reduction, compared to a black-box NN and a variational method. Because of its physical basis, the proposed NN can predict field distributions in rectangular waveguides and the effective indices of higher-order modes.

Islamic Banking Muḍārabah Review from Depositors’ Funds Security Perspective

Islamic banks use Muḍārabah to collect money into their saving and term deposit investment accounts. While doing so these banks stipulate in their terms and conditions that the customers as money providers will solely bear the loss. The banks who manage these funds will neither bear the loss, nor guarantee these funds. Additionally, the customers have no course to the profit distribution or investment decisions. The banks quote that these terms and conditions are in line with Muḍārabah Sharῑ῾ah principles. On the contrary, they use the same terms and conditions as an excuse for not providing financing to the customers based on the Muḍārabah. This double standard draws attention of the researchers to analyze the situation with the aim to protect the depositors in Islamic banks. This is a qualitative study that used unrestricted interview methods to collect data. It was found from research that the current banking Muḍārabah has been altered completely and that the banks don’t qualify to continue using same loss terms and conditions. It is inevitable for the financial and social safety of society to update the Muḍārabah terms and conditions to make them comply with Maqāṣid al Sharῑ῾ah. The study recorded the view of few scholars; however, it opened the doors for Islamic banks, regulators and Fiqh councils to further discuss the legitimacy of the banking Muḍārabah terms and conditions to protect relationship with the customers who deposit money in trust of being Islamic and to avoid the usury offered by the interest based conventional banking.

Interaction of mixed localized waves in optical media with higher-order dispersion

This work focuses on the interaction of mixed localized waves in optical media with higher-order dispersions whose dynamics are governed by a modified cubic–quintic nonlinear Schrödinger equation. For proving the integrability of this model equation, we start by building a Lax pair and an infinitely many conservation laws. Applying the linear stability analysis method, the baseband modulational instability of a stationary continuous wave solution is investigated. Studying the baseband modulational instability phenomenon, we show that the optical loss influences the instability gain spectrum: the stationary continuous wave solution under consideration satisfies the condition of the baseband modulational instability only when the optical loss is neglected. According to the generalized perturbation (n,p−n)–fold Darboux transformation, the existence and properties of the parametric first-, second-, and third-order mixed localized wave solutions for the model equation are constructed when the loss term is neglected. The built solutions helping, we engineer in optical media with higher-order dispersions new nonlinear structures showing interactions between various kinds of nonlinear waves such as multi-peak bright/dark solitons, bright/dark breathers, bright/dark rogue waves, as well as periodic waves. Graphical illustrations are then used for investigating main characteristics of the mixed localized waves propagating on vanishing/nonvanishing continuous wave background. Interestingly, our study produces nonlocal breathers in which the entire optical field oscillates periodically in conjunction with the central local oscillation during transmission. Investigating the effects of various parameters on the nonlinear structures resulting from built mixed localized wave solutions of the model equation, we show that parameter of the fourth-order dispersion can be used to describe wave compression. Also, we show that the model parameters are useful for controlling the optical waves in lossless optical media with both higher-order dispersion whose dynamics are governed by the model equation under consideration. Our results are useful for investigating mixed localized waves in nonlinear metamaterials with cubic–quintic nonlinearity, detuning intermodal dispersion, self steepening and self-frequency effects, and nonlinear third- and fourth-order dispersions.

Synergistic learning with multi-task DeepONet for efficient PDE problem solving.

Multi-task learning (MTL) is an inductive transfer mechanism designed to leverage useful information from multiple tasks to improve generalization performance compared to single-task learning. It has been extensively explored in traditional machine learning to address issues such as data sparsity and overfitting in neural networks. In this work, we apply MTL to problems in science and engineering governed by partial differential equations (PDEs). However, implementing MTL in this context is complex, as it requires task-specific modifications to accommodate various scenarios representing different physical processes. To this end, we present a multi-task deep operator network (MT-DeepONet) to learn solutions across various functional forms of source terms in a PDE and multiple geometries in a single concurrent training session. We introduce modifications in the branch network of the vanilla DeepONet to account for various functional forms of a parameterized coefficient in a PDE. Additionally, we handle parameterized geometries by introducing a binary mask in the branch network and incorporating it into the loss term to improve convergence and generalization to new geometry tasks. Our approach is demonstrated on three benchmark problems: (1) learning different functional forms of the source term in the Fisher equation; (2) learning multiple geometries in a 2D Darcy Flow problem and showcasing better transfer learning capabilities to new geometries; and (3) learning 3D parameterized geometries for a heat transfer problem and demonstrate the ability to predict on new but similar geometries. Our MT-DeepONet framework offers a novel approach to solving PDE problems in engineering and science under a unified umbrella based on synergistic learning that reduces the overall training cost for neural operators.

Economy Wide Impact of Trade Restrictions on Exporters: Evidence Based on a New Measurement Applied to India

Significant trade barriers exist worldwide despite formal efforts at lowering these. Against this backdrop, the present article proposes to examine the economy wide impact of import restrictions on an exporting country based on a new measure for trade restrictiveness. This measure is based on tariffs, non-tariff measures (NTMs) and border trade restrictions imposed by importing countries. Using India as a case, the empirical results of the article suggest that Indian exporters have been grappling with substantial trade restrictions over the years, particularly on account of NTMs. The economy suffered welfare losses due to fall in exports, output, employment, terms of trade and loss in allocative efficiency. Tariffs and NTMs have received much attention from academia and policymakers. But, due to their relatively smaller share in trade restrictions, trading across border restrictions has received much lesser attention. However, as observed in the present article, though border restrictions appear marginal, their welfare and other macroeconomic impacts are substantial. Hence, to ignore these restrictions is likely to result in significant underestimation of trade restrictions faced by exporters. JEL Codes: F1, F13, F140

Deep superpixel generation and clustering for weakly supervised segmentation of brain tumors in MR images

PurposeTraining machine learning models to segment tumors and other anomalies in medical images is an important step for developing diagnostic tools but generally requires manually annotated ground truth segmentations, which necessitates significant time and resources. We aim to develop a pipeline that can be trained using readily accessible binary image-level classification labels, to effectively segment regions of interest without requiring ground truth annotations.MethodsThis work proposes the use of a deep superpixel generation model and a deep superpixel clustering model trained simultaneously to output weakly supervised brain tumor segmentations. The superpixel generation model’s output is selected and clustered together by the superpixel clustering model. Additionally, we train a classifier using binary image-level labels (i.e., labels indicating whether an image contains a tumor), which is used to guide the training by localizing undersegmented seeds as a loss term. The proposed simultaneous use of superpixel generation and clustering models, and the guided localization approach allow for the output weakly supervised tumor segmentations to capture contextual information that is propagated to both models during training, resulting in superpixels that specifically contour the tumors. We evaluate the performance of the pipeline using Dice coefficient and 95% Hausdorff distance (HD95) and compare the performance to state-of-the-art baselines. These baselines include the state-of-the-art weakly supervised segmentation method using both seeds and superpixels (CAM-S), and the Segment Anything Model (SAM).ResultsWe used 2D slices of magnetic resonance brain scans from the Multimodal Brain Tumor Segmentation Challenge (BraTS) 2020 dataset and labels indicating the presence of tumors to train and evaluate the pipeline. On an external test cohort from the BraTS 2023 dataset, our method achieved a mean Dice coefficient of 0.745 and a mean HD95 of 20.8, outperforming all baselines, including CAM-S and SAM, which resulted in mean Dice coefficients of 0.646 and 0.641, and mean HD95 of 21.2 and 27.3, respectively.ConclusionThe proposed combination of deep superpixel generation, deep superpixel clustering, and the incorporation of undersegmented seeds as a loss term improves weakly supervised segmentation.

Self-adaptive weighted physics-informed neural networks for inferring bubble motion in two-phase flow

Modeling complex fluid flow using machine learning is increasingly recognized as a valuable approach for revealing multiphase fluid phenomena. Bubble dynamics represent a classical two-phase flow problem that plays a crucial role in various engineering domains. In this paper, physics-informed neural networks (PINNs) are applied to facilitate incompressible two-phase bubble motion modeling by integrating governing equations and interface evolution equations. The loss function of PINNs consists of multiple loss terms, including initial and boundary conditions constraints, partial differential equations residuals, and volume fraction constraints. The performance of PINNs is influenced by the competing effects of these loss terms. Therefore, we introduce a heuristic adaptive weights approach to automatically adjust loss weights for each training point, avoiding manual tuning and improving the accuracy of PINNs. We investigate typical bubble motion cases, specifically focusing on bubble rising and breakup, to showcase the capabilities of the proposed method. We explore the impact of weights and present the results in comparison to the baselines. Through the bubble breakup case, we illustrate that our model shows superior performance even with more complex scenarios. Then we further discuss the generalization and robustness of our model, showing their indispensability over traditional solvers in gas–liquid two-phase systems. Specifically, we accelerate computation speed in transfer learning without the need to modify the original model. We also show that our method effectively solves ill-posed problems, such as those without initial data or with incomplete or noisy boundary conditions.

Hierarchical existential prior based on expanded pseudo-label for crack detection.

Road crack detection approaches based on the image processing technique have attracted much attention during the past decade due to their convenience and efficiency, but most of them cannot achieve the expected performances due to the complex background interference and severe category imbalance of road images. This paper presents a hierarchical existential prior based on an expanded pseudo-label for crack detection. In particular, the framework contains three variants of U-Net, and each sub-network is trained by pseudo-labels generated by transforming semantic categories of non-crack pixels distributed in the neighborhoods of crack ones. Notably, the expansion degrees of labels for three sub-networks are set in hierarchical descending order. In other words, the crack samples of pseudo-labels for the latter sub-network are a subset of pseudo-labels for the former one, and we define it as an existential prior, which can optimize the network in a coarse-to-fine fashion and refine the detection result gradually. In addition, we utilize a hybrid loss consisting of IoU, SSIM, and focal loss to optimize the network in different aspects, including image-aspect, patch-aspect, and pixel aspect in the training phase, which can improve the structural representation capability of the model. In addition, we present a dynamic hyper-parameter adjustment strategy to balance the weight coefficients of different loss terms, which can enhance the robustness of the model for various practical scenes. Finally, the proposed method achieves 11.36%, 29.76%, and 26.73% in terms of Fβ on CrackTree200, Crack Forest, and ALE datasets, respectively, which sufficiently demonstrate its effectiveness and superiority.

A simple remedy for failure modes in physics informed neural networks.

Physics-informed neural networks (PINNs) have shown promising results in solving a wide range of problems involving partial differential equations (PDEs). Nevertheless, there are several instances of the failure of PINNs when PDEs become more complex. Particularly, when PDE coefficients grow larger or PDEs become increasingly nonlinear, PINNs struggle to converge to the true solution. A noticeable discrepancy emerges in the convergence speed between the PDE loss and the initial/boundary conditions loss, leading to the inability of PINNs to effectively learn the true solutions to these PDEs. In the present work, leveraging the neural tangent kernels (NTKs), we investigate the training dynamics of PINNs. Our theoretical analysis reveals that when PINNs are trained using gradient descent with momentum (GDM), the gap in convergence rates between the two loss terms is significantly reduced, thereby enabling the learning of the exact solution. We also examine why training a model via the Adam optimizer can accelerate the convergence and reduce the effect of the mentioned discrepancy. Our numerical experiments validate that sufficiently wide networks trained with GDM and Adam yield desirable solutions for more complex PDEs.

MVE-Net: A label-free microscopic image visual enhancement network via mRetinex and nonreference loss guidance

Label-free microscopic cell image analysis (segmentation, detection, counting, e.g.) is elementary for unravelling the biological functions of cells and their organelles. However, low contrast, darker brightness, background inhomogeneous, and weak edges of cells cause challenges in subsequent cell image analysis processes. To address these challenges, a Microscopic Visual Enhancement Network (MVE-Net) is proposed to improve microscopic visual effects through pre-enhancement and enhancement processes. In the pre-enhancement stage, to overcome the difficulty of acquiring paired or unpaired images for training, the mRetinex block is proposed to guide the pre-enhancement network to image contrast, cell details, and structural features. Furthermore, a multi-scale extraction module is employed to extract and fuse cell texture and structural features from the pre-enhanced images at various scales, guiding the generator training. In the enhancement stage, a nonreference loss block is designed, incorporating spatial consistency, uneven illumination smoothness, and exposure adjustment loss terms, to further enhance the contrast between cells and the background, smooth the inhomogeneous background, and adjust overall image brightness, thereby guiding the generator's enhancement process and improving the visual effect of microscopic images. Experiments on the LIVECell and PNT1A datasets demonstrate that MVE-Net outperforms state-of-the-art image enhancement methods, significantly improving image contrast, brightness, cell detail, and structural features without the need for paired or unpaired reference standard images for training.

Lifetime Prolongation of PtCo/C Cathode Catalysts for PEM Fuel Cells By Understanding Triggers for Pt Surface Area Loss and Co Dissolution in Voltage Cycling-Based Accelerated Stress Tests

Proton exchange membrane fuel cell (PEMFC) applications in the heavy-duty mobility sector have attracted increasing attention due to their flexible scalability in terms of energy and power density. The challenging high fuel efficiency targets in the heavy-duty sector can be met by introducing platinum-cobalt alloy (PtxCo/C) catalysts which exhibit a 1.5-3x higher kinetic activity for the oxygen reduction reaction (ORR) as compared to carbon supported platinum (Pt/C) catalysts.1 Despite the better beginning-of-life performance of PtxCo/C, the dissolution of Co2+ ions into the ionomer phase of the membrane electrode assembly (MEA) during aging results in significant performance losses.2 The release of Co2+ ions from the catalyst influences the fuel cell performance in two major ways: i) a decrease in specific ORR activity, most prominent at low-current densities, and thus in fuel cell efficiency, and ii) mass transport-related performance losses due to Co2+ cation contamination of the ionomer phase, being especially relevant at high-current densities.2 Besides increasing the lifetime by using advanced catalysts (e.g., core-shell particles), longer durability can be achieved by optimizing the fuel cell operation conditions to minimize Pt surface area loss and Co dissolution.3 In this study, we investigate the triggers for the performance decay of PtxCo/C ORR catalysts by separately evaluating Pt and Co loss by varying the conditions of voltage cycling accelerated stress tests (ASTs). All voltage cycling ASTs are performed using 5 cm2 active area single-cells with membrane electrode assemblies (MEAs) based on PtxCo/C cathode catalysts at a loading of 0.3 mgPt cmMEA -2, conducted in H2/N2 at 80 °C, 95% RH, and ambient pressure. Whereas the lower potential limit (LPL) is kept constant for all tests, different upper potential limits (UPLs), UPL hold times, and wave profiles are investigated. The MEA performance at different aging steps (i.e., number of voltage cycles) is evaluated as a function of Pt ECSA (electrochemically active surface area) loss and Co dissolution. Whereas previous voltage cycling studies for Pt/C catalysts revealed that the H2/air performance losses are solely determined by the cathode roughness factor (rf, in cm2 Pt/cm2 cathode), independent of the aging protocol,4 we show that this relation does not hold for PtxCo/C catalysts. Instead, the H2/air performance needs to additionally account for the extent of Co dissolution. Both loss terms are partly independent and can be triggered to varying extents depending on the aging protocol. Based on our findings, we propose guidelines for possible operation strategies to minimize PtxCo/C degradation under automotive conditions. References C. Lim, A. R. Fairhurst, B. J. Ransom, D. Haering, V. R. Stamenkovic, ACS Catalysis 2023, 13, 14874-14893.N. Ramaswamy, S. Kumaraguru, W. Gu, R. S. Kukreja, K. Yu, D. Groom, P. Ferreira, Journal of The Electrochemical Society 2021, 168, 024519.N. Ramaswamy, S. Kumaraguru, R. S. Kukreja, D. Groom, K. Jarvis, P. Ferreira, Journal of The Electrochemical Society 2021, 168, 124512.R. K. F. Della Bella, B. M. Stühmeier, H. A. Gasteiger, Journal of The Electrochemical Society 2022, 169, 044528. Acknowledgement This research has received funding from Bosch.

Physics-Informed Neural Networks for State and Parameter Estimation of Li-Ion Batteries

Combining machine learning with physics is a trending approach for discovering unknown dynamics, and one of the most intensively studied frameworks is the physics-informed neural network (PINN). However, PINN often fails to optimize the network due to its difficulty in concurrently minimizing multiple losses originating from the system’s governing equations. This problem can be more serious when the system’s states are unmeasurable, like lithium-ion batteries. In this work, we introduce a robust method for training PINN that uses fewer loss terms and thus constructs a less complex landscape for optimization. In particular, instead of having loss terms from each differential equation, this method embeds the dynamics into a loss function that quantifies the error between observed and predicted system outputs. This is accomplished by numerically integrating the predicted states from the neural network(NN) using known dynamics and transforming them to obtain a sequence of predicted outputs. Minimizing such a loss optimizes the NN to predict states consistent with observations given the physics. Further, the system’s parameters can be added to the optimization targets. To demonstrate the ability of this method to perform various modeling and control tasks, we apply it to an electrochemical battery model to concurrently estimate its states and parameters. Figure 1

Enhancing Proton Exchange Membrane Fuel Cell Durability By Using Porous Carbon-Supported Cathode Catalysts with Various Pt-Particle Sizes

Nowadays, proton exchange membrane fuel cells (PEMFCs) are envisioned for heavy-duty vehicles (HDVs) such as trucks and buses, due to their high power density, low emissions, and quiet operation.1 However, to be viable for HDVs, PEMFCs require cathode catalyst materials that are highly stable and durable in order to maintain performance over at least 30,000 hours of operation. One well-known challenge is that voltage cycling of the cathode leads to significant losses in the electrochemically active surface area (ECSA) due to Pt-dissolution and subsequent growth via Ostwald ripening or particle coalescence, as well as Pt loss into the ionomer/membrane phase.2 While larger Pt particles are less prone to dissolution due to their lower surface free energy,3 depositing Pt particles within the internal pores of primary porous carbon particles has been shown to effectively mitigate particle coalescence.4 Therefore, combining larger particles with locating them internally of the carbon pores shows promise for significantly reducing Pt surface area loss.In this study, voltage cycling-based accelerated stress tests (ASTs) were performed using 5 cm2 membrane electrode assemblies (MEAs) with 0.1 mgPt cmMEA -2 loaded Pt-based cathode catalysts supported on porous Ketjenblack carbon (TEC10E20E, TKK, referred to as “pristine” Pt/KB). To increase the Pt-particle size, the catalyst was subjected to a heat-treatment under an reductive atmosphere (5% H2/Ar mixture) at 900 °C for 1 h (referred to as “HT-900°C/1h”), while a subsequent treatment under oxidative atmosphere (under air/Ar mixture) at 250 °C for 12 h (referred to as “ox-HT-900°C/1h”) was also included to enhance catalyst accessibility by inducing a Pt-catalyzed local carbon support oxidation of the bottleneck pore entrances, as described by Lazaridis et al.5 A full MEA characterization was carried out, encompassing determination of the ECSA by CO-stripping, of the H2/air and H2/O2 performance, and of the O2 and H+ transport resistances in the cathode by limiting current measurements and electrochemical impedance spectroscopy, respectively.We found that the HT-900°C/1h cathode exhibited significantly improved ECSA retention compared to the pristine Pt/KB, primarily due to its larger Pt particle size (~5 vs. ~2.5 nm), which reduces Pt dissolution. However, it was also observed that these catalysts had a much lower beginning-of-life (BoL) H2/air performance, as evidenced by comparing the rightmost data points in Fig. 1 for the pristine (gray squares) and the HT-900°C/1h (blue triangles) catalysts. These performance losses are attributed to the reduced roughness factor (rf), i.e., the lower available Pt surface area per electrode area, which was demonstrated to affect various voltage loss terms.6 Especially at higher current densities, an increased impact of the local oxygen transport resistance is observed, due to its inverse proportionality to the cathode rf. To address these challenges, it was demonstrated that enhancing catalyst accessibility substantially improves H2/air performance at higher current densities, so that the ox-HT-900°C/1h catalyst surpasses the initial performance of the pristine Pt/KB (compare rightmost green circles). Thus, combining heat treatment of a Pt/KB catalyst under both reductive and oxidative atmosphere enhances cathode catalyst durability and concomitantly high current density H2/air performance.

A deep neural network to de-noise single-cell RNA sequencing data.

Single-cell RNA sequencing (scRNA-seq), a powerful technique for investigating the transcriptome of individual cells, enables the discovery of heterogeneous cell populations, rare cell types, and transcriptional dynamics in separate cells. Yet, scRNA-seq data analysis is limited by the problem of measurement dropouts, i.e., genes displaying zero expression levels. We introduce ZiPo, a deep artificial neural network for rate estimation and library size prediction in scRNA-seq data which incorporates adjustable zero inflation in the distribution to capture the dropouts. ZiPo builds upon established concepts, including using deep autoencoders and adopting the Poisson and negative binomial distributions, by taking advantage of novel strategies, including library size prediction and residual connections, to improve the overall performance. A significant innovation of ZiPo is the introduction of a scale-invariant loss term, making the weights sparse and, hence, the model biologically more interpretable. ZiPo quickly handles vast singular and mixed datasets, with the processing time directly proportional to the number of cells. In this paper, we demonstrate the power of ZiPo on three datasets and show its advantages over other current techniques. The code used to produce the results in this manuscript is available at https://bitbucket.org/habilzare/alzheimer/src/master/code/deep/ZiPo/.

Temporal attention fusion network with custom loss function for EEG–fNIRS classification

Objective.Methods that can detect brain activities accurately are crucial owing to the increasing prevalence of neurological disorders. In this context, a combination of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) offers a powerful approach to understanding normal and pathological brain functions, thereby overcoming the limitations of each modality, such as susceptibility to artifacts of EEG and limited temporal resolution of fNIRS. However, challenges such as class imbalance and inter-class variability within multisubject data hinder their full potential.Approach.To address this issue, we propose a novel temporal attention fusion network (TAFN) with a custom loss function. The TAFN model incorporates attention mechanisms to its long short-term memory and temporal convolutional layers to accurately capture spatial and temporal dependencies in the EEG-fNIRS data. The custom loss function combines class weights and asymmetric loss terms to ensure the precise classification of cognitive and motor intentions, along with addressing class imbalance issues.Main results.Rigorous testing demonstrated the exceptional cross-subject accuracy of the TAFN, exceeding 99% for cognitive tasks and 97% for motor imagery (MI) tasks. Additionally, the ability of the model to detect subtle differences in epilepsy was analyzed using scalp topography in MI tasks.Significance.This study presents a technique that outperforms traditional methods for detecting high-precision brain activity with subtle differences in the associated patterns. This makes it a promising tool for applications such as epilepsy and seizure detection, in which discerning subtle pattern differences is of paramount importance.

Latent Graph Attention for Spatial Context in Light-Weight Networks: Multi-Domain Applications in Visual Perception Tasks

Global contexts in images are quite valuable in image-to-image translation problems. Conventional attention-based and graph-based models capture the global context to a large extent; however, these are computationally expensive. Moreover, existing approaches are limited to only learning the pairwise semantic relation between any two points in the image. In this paper, we present Latent Graph Attention (LGA), a computationally inexpensive (linear to the number of nodes) and stable modular framework for incorporating the global context in existing architectures. This framework particularly empowers small-scale architectures to achieve performance closer to that of large architectures, making the light-weight architectures more useful for edge devices with lower compute power and lower energy needs. LGA propagates information spatially using a network of locally connected graphs, thereby facilitating the construction of a semantically coherent relation between any two spatially distant points that also takes into account the influence of the intermediate pixels. Moreover, the depth of the graph network can be used to adapt the extent of contextual spread to the target dataset, thereby able to explicitly control the added computational cost. To enhance the learning mechanism of LGA, we also introduce a novel contrastive loss term that helps our LGA module to couple well with the original architecture at the expense of minimal additional computational load. We show that incorporating LGA improves performance in three challenging applications, namely transparent object segmentation, image restoration for dehazing and optical flow estimation.

Pipeline leak detection based on generative adversarial networks under small samples

During the actual industrial process, it is challenging to obtain samples of oil and gas pipeline leaks resulting in the scarcity of training data samples suitable for fault diagnosis. In order to tackle this issue, this paper suggests a method a Recursive Generalization self-attention generative adversarial network (RAGAN) aided by Wasserstein distance with gradient penalty. The initial step involves applying the short time Fourier transform to the acoustic signal of oil and gas pipeline leakage, treating it as real data. Subsequently, both the real data and random noise following a Gaussian distribution are fed into the generator. The output is utilised as a pseudo sample. The Wasserstein distance of the distribution of real data and fake samples is introduced as a loss term in the discriminator, and a gradient penalty is added. Finally, the network optimizes the parameters through back propagation until Nash equilibrium. PSNR and SSIM are used as sample reliability evaluation. The results show that the fake samples have high similarity with the real samples, which can be used to expand small sample data. Moreover, extending pseudo samples to small sample data sets can effectively improve the performance of fault diagnosis.

Recurrent neural networks as a physics-based self-learning solver to satisfy plane stress viscoplasticity undergoing isotropic damage

The present study introduces a new self-learning framework based on a recurrent neural network to compute the nonlinear viscoplastic response of plate structures undergoing isotropic damage. The neural network comprising Legendre Memory Unit (LMU) cells and dense transformations is pretrained with a combination of data-driven and physics-based loss functions. Once pretrained, the model is deployed in the explicit integration scheme to predict the strain increment in the thickness direction satisfying plane stress condition. This results in elimination/reduction in the iterations required to fulfill the plane stress condition and leads to faster convergence since the neural network replaces the root-finding methods. The proposed self-learning method differentiates itself from the well-known Physics-Informed Neural Networks (PINNs) since the neural network model self-learns online through the proposed physics-based loss term without the need for data. We compare the results of the Neural Network-enhanced integration scheme to validate the proposed method with those of the classical explicit integration scheme.

Learning Gravity Fields of Small Bodies: Self-adaptive Physics-informed Neural Networks

The reconstruction of the gravity field within the surface region of small bodies is crucial for the surface proximity operations of a probe. However, the irregular shape, uneven mass distribution, and sparse gravitational data of small bodies pose challenges in the reconstruction. We propose a self-adaptive physics-informed neural network (PINN) for the reconstruction of the gravity field within the surface region of irregular and heterogeneous small bodies. First, we introduce an auxiliary-point-based data augmentation strategy to reduce the model’s dependency on the quantity of data. Second, we incorporate a residual-based adaptive sampling strategy to enhance the prediction accuracy of the model in regions with significant variations in small-body density. Finally, we introduce an adaptive weight module based on gradient ascent to mitigate the balancing issue of loss terms in the PINN. Experiments indicate that our algorithm achieves improved accuracy for reconstructing the gravity field within the surface region of small bodies. This work is expected to contribute to the enhancement of safety in surface proximity operations around the surfaces of small bodies.

Multi-channel neural audio decorrelation using generative adversarial networks

The degree of correlation between the sounds received by the ears significantly influences the spatial perception of a sound image. Audio signal decorrelation is, therefore, a commonly used tool in various spatial audio rendering applications. In this paper, we propose a multi-channel extension of a previously proposed decorrelation method based on generative adversarial networks. A separate generator network is employed for each output channel. All generator networks are optimized jointly to obtain a multi-channel output signal with the desired properties. The training objective includes a number of individual loss terms to control both the input-output and the inter-channel correlation as well as the quality of the individual output channels. The proposed approach is trained on music signals and evaluated both objectively and through formal listening tests. Thereby, a comparison with two classical signal processing-based multi-channel decorrelators is performed. Additionally, the influence of the number of output channels, the individual loss term weightings, and the employed training data on the proposed method’s performance is investigated.