Auxiliary Datasets Research Articles

Online learning discriminative sparse convolution networks for robust UAV object tracking

Despite the remarkable empirical success for UAV object tracking, current convolutional networks usually have three unavoidable limitations: (1) The feature maps produced by convolutional layers are difficult to interpret. (2) The network needs to be trained offline on a large-scale auxiliary training set, resulting in the feature extraction ability of the trained network depending on the categories of the training set. (3) The performance of networks suffers from sensitivity to hyper-parameters (such as learning rate and weight decay) when the network needs online fine-tuning. To overcome the three limitations, this paper proposes a Discriminative Sparse Convolutional Network (DSCN) that exhibits good layer-wise interpretability and can be trained online without requiring any auxiliary training data. By imposing sparsity constraints on the convolutional kernels, DSCN furnishes the convolution layer with an explicit data meaning, thus enhancing the interpretability of the feature maps. These convolutional kernels are directly learned online from image blocks, which eliminates the offline training process on auxiliary data sets. Moreover, a simple yet effective online tuning method with few hyper-parameters is proposed to fine-tune fully connected layers online. We have successfully applied DSCN to UAV object tracking and conducted extensive experiments on six mainstream UAV datasets. The experimental results demonstrate that our method performs favorably against several state-of-the-art tracking algorithms in terms of tracking accuracy and robustness.

Full-coverage estimation of CO2 concentrations in China via multisource satellite data and Deep Forest model

Monitoring China’s carbon dioxide (CO2) concentration is essential for formulating effective carbon cycle policies to achieve carbon peaking and neutrality. Despite insufficient satellite observation coverage, this study utilizes high-resolution spatiotemporal data from the Orbiting Carbon Observatory 2 (OCO-2), supplemented with various auxiliary datasets, to estimate full-coverage, monthly, column-averaged carbon dioxide (XCO2) values across China from 2015 to 2022 at a spatial resolution of 0.05° via the deep forest model. The 10-fold cross-validation results indicate a correlation coefficient (R) of 0.95 and a determination coefficient (R²) of 0.90. Validation against ground-based station data yielded R values of 0.93, and R² values reached 0.81. Further validation from the Greenhouse Gases Observing Satellite (GOSAT) and the Copernicus Atmosphere Monitoring Service Reanalysis dataset (CAMS) produced R² values of 0.87 and 0.80, respectively. During the study period, CO2 concentrations in China were higher in spring and winter than in summer and autumn, indicating a clear annual increase. The estimates generated by this study could potentially support CO2 monitoring in China.

Scalability challenges of machine learning models for estimating walking and cycling volumes in large networks

This study explores the scalability of machine learning models for estimating walking and cycling volumes across the extensive New South Wales (NSW) Six Cities Region in Australia using mobile phone and crowdsourced data. Previous research has focused on localized applications, missing the complexities of larger networks. The research addresses this gap by identifying unique challenges such as the scarcity and representativeness of observed count data, gaps in the crowdsourced and mobile phone data, and inconsistencies in link-level volume estimates. We propose and demonstrate the application of strategies like enhancing geographical diversity of observed count data and employing an extensive cross-validation approach in model training and testing. By leveraging various auxiliary datasets, the study demonstrates the effectiveness of these strategies in improving model performance. These findings provide valuable insights for transportation modelers, policymakers, and urban planners, offering a robust framework for supporting sustainable transportation infrastructure and policies with advanced data-driven methodologies.

A New Joint Training Method for Facial Expression Recognition with Inconsistently Annotated and Imbalanced Data

Facial expression recognition (FER) plays a crucial role in various applications, including human–computer interaction and affective computing. However, the joint training of an FER network with multiple datasets is a promising strategy to enhance its performance. Nevertheless, widespread annotation inconsistencies and class imbalances among FER datasets pose significant challenges to this approach. This paper proposes a new multi-dataset joint training method, Sample Selection and Paired Augmentation Joint Training (SSPA-JT), to address these challenges. SSPA-JT models annotation inconsistency as a label noise problem and selects clean samples from auxiliary datasets to expand the overall dataset size while maintaining consistent annotation standards. Additionally, a dynamic matching algorithm is developed to pair clean samples of the tail class with noisy samples, which enriches the tail classes with diverse background information. Experimental results demonstrate that SSPA-JT achieved superior or comparable performance compared with the existing methods by addressing both annotation inconsistencies and class imbalance during multi-dataset joint training. It achieved state-of-the-art performance on RAF-DB and CAER-S datasets with accuracies of 92.44% and 98.22%, respectively, reflecting improvements of 0.2% and 3.65% over existing methods.

A Zero Auxiliary Knowledge Membership Inference Attack on Aggregate Location Data

Location data is frequently collected from populations and shared in aggregate form to guide policy and decision making. However, the prevalence of aggregated data also raises the privacy concern of membership inference attacks (MIAs). MIAs infer whether an individual's data contributed to the aggregate release. Although effective MIAs have been developed for aggregate location data, these require access to an extensive auxiliary dataset of individual traces over the same locations, which are collected from a similar population. This assumption is often impractical given common privacy practices surrounding location data. To measure the risk of an MIA performed by a realistic adversary, we develop the first Zero Auxiliary Knowledge (ZK) MIA on aggregate location data, which eliminates the need for an auxiliary dataset of real individual traces. Instead, we develop a novel synthetic approach, such that suitable synthetic traces are generated from the released aggregate. We also develop methods to correct for bias and noise, to show that our synthetic-based attack is still applicable when privacy mechanisms are applied prior to release. Using two large-scale location datasets, we demonstrate that our ZK MIA matches the state-of-the-art Knock-Knock (KK) MIA across a wide range of settings, including popular implementations of differential privacy (DP) and suppression of small counts. Furthermore, we show that ZK MIA remains highly effective even when the adversary only knows a small fraction (10%) of their target's location history. This demonstrates that effective MIAs can be performed by realistic adversaries, highlighting the need for strong DP protection.

Bearing RUL prediction and fault diagnosis system based on parallel multi-scale MIMT lightweight model

Abstract In actual industrial production, the importance of safety production is increasingly prominent, and the degradation and failure of machinery and equipment are potential sources of safety hazards. Therefore, there is a growing trend towards real-time monitoring, prediction, and diagnosis of industrial equipment to prevent unpredictable impacts on life and property safety caused by sudden failures. To address this issue, this paper proposes a real-time degradation anomaly detection based on parallel multiscale autoencoders and a lightweight model of parallel multiscale multi-input multi-task for bearing Remaining Useful Life (RUL) prediction and fault diagnosis systems. Firstly, the multiscale autoencoder method is used to simulate actual working conditions and reconstruct the original vibration signals to build abnormal degradation detection intervals. The [0, $\mu$ +3$\sigma$] interval is utilized to judge abnormal degradation based on reconstruction errors, and the First Predict Timepoint (FPT) is determined adaptively. Secondly, a method for constructing dimensionless auxiliary datasets is proposed, which adopts a multi-input form based on deep separable convolution for feature extraction of original vibration signals, kurtosis, and peak values to improve the prediction and diagnosis performance of the lightweight model. Finally, a multi-task output method combining clustering and regression is employed to achieve RUL prediction and fault diagnosis of bearings. The proposed method overcomes the problems existing in traditional bearing RUL prediction and diagnosis methods and possesses theoretical innovation and engineering practicality. Validation on two bearing datasets confirms the effectiveness of the proposed method.

Few-shot remote sensing image segmentation based on label propagation and open-set domain adaptation

ABSTRACT Few-shot segmentation aims to segment objects in a task dataset with only a few labelled target examples, aided by a lot of labelled auxiliary datasets. Existing approaches typically assume that the task dataset and auxiliary dataset originate from the same source but have disjoint category sets. However, it is common in remote sensing image segmentation tasks that the datasets are from different sources and the category sets are often partially overlapped, where models trained on the auxiliary dataset degrade in performance on the task dataset. To address these issues, we present a few-shot segmentation method based on graph network and open-set domain adaptation. The proposed method utilizes a graph model to capture the relationships between super pixels and to predict labels for the unlabelled samples by label propagation, meanwhile, leverages open-set domain adaptation to reduce inter-domain discrepancies and enhances model transferability. The proposed method is evaluated on two publicly available datasets, and the effectiveness of the proposed approach is demonstrated by the significant improvement in accuracy compared to existing methods.

Advancements in mapping areas suitable for wetland habitats across the conterminous United States

Wetland habitats provide critical ecosystem services to the surrounding landscape, including nutrient and pollutant retention, flood mitigation, and carbon storage. Wetland connectivity to water bodies and related ecosystems is critical in habitat sustainability, but there are limited resources for landscape-level wetland planning. Considering the network connectivity of an ecosystem type can derive different benefits to the natural and built environment, as well as human health. The value that wetlands provide, along with incentive programs and conservation goals mandated by the government require new and improved wetland spatial data. Utilizing high quality, publicly available data, this study finds that the amount of land in the United States that could support built or restored wetlands is more than double the area of mapped existing wetlands.This study uses 17 input variables (i.e., features extracted from remotely sensed data and auxiliary datasets) at the 10-m resolution and the National Wetlands Inventory to train a random forest model to identify areas that may support a wetland habitat, or potential wetland areas. Models were calculated for each of 18 two-digit hydrologic units that encompass the conterminous United States, and model overall accuracy ranged from 78.0 % to 89.8 %. The models predicted that 21.1 % of the conterminous United States can be categorized as potential wetland area.Selecting input variables to predict areas with wetland potential, rather than to identify existing wetlands, using the random forest algorithm can be transferred to other locations, scales, and ecosystem types. Visualizing potential wetland areas using input data at the 10-m resolution and enhanced methodology improves previous work, as even slight changes in topography, soils, and landscape features can determine ecosystem connections. This product can be used to better place wetland restoration projects to serve ecosystem- and community-wide health by ensuring ecosystem success and targeting areas that face increased climate change impacts.

Challenges in applying water budget framework for estimating groundwater storage changes from GRACE observations

The application of a water budget framework to isolate Gravity Recovery and Climate Experiment (GRACE) groundwater storage anomalies (GRACE-GWA) from GRACE terrestrial water storage anomalies (GRACE-TWSA) is hindered by the lack of direct observations of water budget components. In GRACE groundwater studies, water budget components are frequently applied to isolate changes in a storage component from various auxiliary methods (e.g., land surface or hydrology models, reanalysis, or remote sensing) and are used as a supplement to, or a substitute for, in-situ measurements when direct observations are sparse or unavailable. The contribution of select auxiliary datasets to isolate GRACE-GWA from GRACE-TWSA is an enduring quandary, attributed to the various assumptions applied to resemble hydrologic processes in model formulations. This study systematically allocates water budget components from assorted auxiliary sources to demonstrate bias and distortion of GRACE-GWA resulting simply from water storage component data selection. Whereas previous GRACE-GWA studies used combinations of water budget component datasets with uncertainty derived from the variance of the combined water budget components, this study applies single estimates for each component to measure bias and variability. An initial comparative analysis focused on three basins with suitable in-situ groundwater observations and complex hydrology stores (e.g., large lakes and seasonal snow cover), with correlation coefficients ranging from 0.32 to 0.89. Our systematic analysis was extended to 12 additional basins that capture a range of hydrologic characteristics to highlight the inconsistency in GRACE-GWA estimates. The variability evident in GRACE-GWA estimates highlights that the selection of water budget components carries the risk of misleading outcomes when disaggregating GRACE-TWSA into GRACE-GWA. Our water budget intercomparison provides an important assessment of the limitation in extraction of GRACE-GWA, highlighting the usefulness of comprehensive appraisal of water budget components for GRACE groundwater studies.

Aligning the domains in cross domain model inversion attack

Model Inversion Attack reconstructs confidential training dataset from a target deep learning model. Most of the existing methods assume the adversary has an auxiliary dataset that has similar distribution with the private dataset. However, this assumption does not always hold in real-world scenarios. Since the private dataset is unknown, the domain divergence between the auxiliary dataset and the private dataset is inevitable. In this paper, we use Cross Domain Model Inversion Attack to represent the distribution divergence scenario in MIA. With the distribution divergence between the private images and auxiliary images, the distribution between the feature vectors of the private images and those of the auxiliary images is also different. Moreover, the outputted prediction vectors of the auxiliary images are also misclassified. The inversion attack is thus hard to be performed. We perform both the feature vector inversion task and prediction vector inversion task in this cross domain setting. For feature vector inversion, Domain Alignment MIA (DA-MIA) is proposed. While performing the reconstruction task, DA-MIA aligns the feature vectors of auxiliary images with the feature vectors of private images in an adversarial manner to mitigate the domain divergence between them. Thus, semantically meaningful images can be reconstructed. For prediction vector inversion, we further introduce an auxiliary classifier and propose Domain Alignment MIA with Auxiliary Classifier (DA-MIA-AC). The auxiliary classifier is pretrained by the auxiliary dataset and fine-tuned during the adversarial training stage. Thus, the misclassification problem caused by domain divergence can be solved, and the images can be reconstructed correctly. Various experiments are performed to show the advancement of our methods, the results show that DA-MIA can improve the SSIM score of the reconstructed images for up to 191%, DA-MIA-AC can increase the classification accuracy score of the reconstructed images from 9.18% to 81.32% in Cross Domain Model Inversion Attack.

Ribonanza: deep learning of RNA structure through dual crowdsourcing.

Prediction of RNA structure from sequence remains an unsolved problem, and progress has been slowed by a paucity of experimental data. Here, we present Ribonanza, a dataset of chemical mapping measurements on two million diverse RNA sequences collected through Eterna and other crowdsourced initiatives. Ribonanza measurements enabled solicitation, training, and prospective evaluation of diverse deep neural networks through a Kaggle challenge, followed by distillation into a single, self-contained model called RibonanzaNet. When fine tuned on auxiliary datasets, RibonanzaNet achieves state-of-the-art performance in modeling experimental sequence dropout, RNA hydrolytic degradation, and RNA secondary structure, with implications for modeling RNA tertiary structure.

Comparison of different machine-learning algorithms for land use land cover mapping in a heterogenous landscape over the Eastern Nile river basin, Ethiopia

Land use/land cover (LULC) information is regarded as one of the most important variables in global change studies. Several LULC classification algorithms are available with different levels of accuracy in LULC mapping under different geographical setups. Furthermore, different remote sensing (RS) based indices and auxiliary data have been reported to have contributed to the accuracy of LULC mapping. Hence, the main aim of this study was to compare the different ML algorithms for the LULC mapping and explicitly assess the potential contribution of different RS-derived indices and auxiliary data on classification accuracy in the Upper Tekeze-Atbarah (UTA) river basin in Ethiopia. The overall classifiers evaluation result revealed that random forest (RF) performed best and followed by K-Nearest Neighbour (KNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), Maximum Likelihood Classification (MLC), and Classification and Regression Tree (CART). Compared to RF, the MLC method showed less accuracy by 16.2% (19.7%) in OA (kappa coefficient) respectively, which showed that the currently emerging ML classifiers contributed to an improvement in LULC mapping. The use of RS-derived indices and DEM as predictors together with the Landsat bands improved the overall accuracy and kappa coefficient by 5.5% and 7% respectively. Furthermore, the discrepancy in the classified LULC map from RF and other classifiers was compared at pixel-level and the result showed a considerable disagreement between the LULC maps. For instance, when the LULC map from RF and KNN are compared, only 80.6% of the area has the same LULC condition. This result reflects the potential discrepancy that could exist in LULC maps from different classifiers. The hydrological simulation under each of the classified map also resulted in a considerable difference in the major water balance components. In a conclusion, given the considerable difference in classifier performance and the potential propagation of errors in LULC maps to further applications such as water resources assessment, it is highly recommended to explicitly evaluate the available classifiers. Users should also harness the power of RS indices and auxiliary datasets for improved LULC mapping.

Multi-source deep domain adaptation ensemble framework for cross-dataset motor imagery EEG transfer learning

Objective. Electroencephalography (EEG) is an important kind of bioelectric signal for measuring physiological activities of the brain, and motor imagery (MI) EEG has significant clinical application prospects. Convolutional neural network has become a mainstream algorithm for MI EEG classification, however lack of subject-specific data considerably restricts its decoding accuracy and generalization performance. To address this challenge, a novel transfer learning (TL) framework using auxiliary dataset to improve the MI EEG classification performance of target subject is proposed in this paper. Approach. We developed a multi-source deep domain adaptation ensemble framework (MSDDAEF) for cross-dataset MI EEG decoding. The proposed MSDDAEF comprises three main components: model pre-training, deep domain adaptation, and multi-source ensemble. Moreover, for each component, different designs were examined to verify the robustness of MSDDAEF. Main results. Bidirectional validation experiments were performed on two large public MI EEG datasets (openBMI and GIST). The highest average classification accuracy of MSDDAEF reaches 74.28% when openBMI serves as target dataset and GIST serves as source dataset. While the highest average classification accuracy of MSDDAEF is 69.85% when GIST serves as target dataset and openBMI serves as source dataset. In addition, the classification performance of MSDDAEF surpasses several well-established studies and state-of-the-art algorithms. Significance. The results of this study show that cross-dataset TL is feasible for left/right-hand MI EEG decoding, and further indicate that MSDDAEF is a promising solution for addressing MI EEG cross-dataset variability.

Assessing the impact of agriculture, coal mining, and ecological restoration on water sustainability in the Mu Us Sandyland

Balancing water demand for socio–economic development and ecosystem stability presents a challenge for regional sustainable management, especially in drylands. Previous studies have indicated that large–scale ecological restoration projects (ERPs) lead to a decline in terrestrial water storage (TWS) in the Mu Us Sandyland (MUS). However, the effects of other human activities (e.g., cropland reclamation, coal mining) on water resources remain unclear, raising concerns regarding water crisis and human–natural system sustainability. Through the utilization of coal mine location data, we found that the impact of coal mass loss on the Gravity Recovery and Climate Experiment (GRACE) products cannot be ignored in MUS, especially in the coal–rich northeastern part. Combining these data with auxiliary datasets, we observed a significant (p < 0.05) decrease in TWS (−0.85 cm yr−1) and groundwater storage (GWS, −0.95 cm yr−1) in the MUS, with human activities accounting for 79.23 % of TWS and 90.45 % of GWS reductions, primarily due to increased agricultural and industrial water consumption. Agricultural water consumption increased 2.23 times from 2001 to 2020, attributed to enhanced water use intensity (62.6 %) and cropland expansion (37.4 %). Industrial water consumption in Shenmu, a representative coal county, experienced a 4.16–fold rise between 2001 and 2020. Despite these challenges, local governments have alleviated water stress, ensured food security, and increased household income by comprehensive management strategies, such as enhancing water–saving technology and enforcing stringent policies. Previous studies have overestimated the amount of water resources consumed by ERPs. However, ERPs has played a critical role in stabilizing the regional ecological environment and ensuring the region as a vital food and energy supplier. Our findings can guide for socio-economic development and water management policies in similar regions.

A review of the BuFeng-1 GNSS-R mission: calibration and validation results of sea surface and land surface

ABSTRACT In this paper, we will conclude the results of Bufeng-1 (BF-1) A/B data processing, calibration workflow, and validation of the calibrated sea surface winds, land surface soil moisture, and sea surface height measurements. Since 2019, the BF-1 mission has operated in-orbit for over 4 years. The Earth reflected delay Doppler maps (DDMs) are continuously collected to perform global sea surface and land observations. At the same time, the intermediate frequency (IF) raw data are also obtained for 12 seconds every pass in diagnostic mode. To begin with, a brief description of the spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) technique will be provided in the introduction. Next, we will present the overview of Chinese BF-1 mission and the data specifications used in our research. In the next section, the BF-1 mission-related spaceborne power calibration and validation are presented to show the support to power DDM observable production for sea surface and land surface applications. Then, the status of Chinese Beidou System (BDS) Equivalent Isotropic Radiated Power (EIRP) acquisition programme is then introduced. Furthermore, the latest sea surface height (SSH) measurements results including two modes (group delay and carrier phase) and wind speed derivation based on machine learning (ML) method will be spatial-temporal aligned and validated with auxiliary datasets including Denmark Technology University (DTU) mean sea surface (MSS) products and European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis. The previous published results of sea surface winds retrieval under Hurricane conditions and soil moisture retrieval are also reviewed for the BF-1 mission applications. Finally, the conclusion of BF-1 derived results will be discussed to draw out ongoing/future works.

An improved OMI ozone profile research product version 2.0 with collection 4 L1b data and algorithm updates

Abstract. We describe the new and improved version 2 of the ozone profile research product from the Ozone Monitoring Instrument (OMI) on the Aura satellite. One of the major changes is to switch the OMI L1b data from collection 3 to the recent collection 4 as well as the accompanying auxiliary datasets. The algorithm details are updated on radiative transfer model calculation and measurement calibrations, along with the input changes in meteorological data, and with the use of a tropopause-based ozone profile climatology, an improved high-resolution solar reference spectrum, and a recent ozone absorption cross-section dataset. A super Gaussian is applied to better represent OMI slit functions instead of a normal Gaussian. The effect of slit function errors on the spectral residuals is further accounted for as pseudo-absorbers in the iterative fit process. The OMI irradiances are averaged into monthly composites to reduce noise uncertainties in OMI daily measurements and to cancel out the temporal variations of instrument characteristics that are common in both radiance and irradiance measurements, which was previously neglected due to use of climatological composites. The empirical soft calibration spectra are re-derived to be consistent with the updated implementations and derived annually to remove the time-varying systematic biases between measured and simulated radiances. The “common mode” correction spectra are derived from remaining residual spectra after soft calibration as a function of solar zenith angle. The common mode is included as a pseudo-absorber in the iterative fit process, which helps to reduce the discrepancies of ozone retrieval accuracy between lower and higher solar zenith angles and between nadir and off-nadir pixels. Validation with ozonesonde measurements demonstrates the improvements of ozone profile retrievals in the troposphere, especially around the tropopause. The retrieval quality of tropospheric column ozone is improved with respect to the seasonal consistency between winter and summer as well as the long-term consistency before and after the row-anomaly occurrence.

Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection

Recent unsupervised anomaly detection methods often rely on feature extractors pretrained with auxiliary datasets or on well-crafted anomaly-simulated samples. However, this might limit their adaptability to an increasing set of anomaly detection tasks due to the priors in the selection of auxiliary datasets or the strategy of anomaly simulation. To tackle this challenge, we first introduce a prior-less anomaly generation paradigm and subsequently develop an innovative unsupervised anomaly detection framework named GRAD, grounded in this paradigm. GRAD comprises three essential components: (1) a diffusion model (PatchDiff) to generate contrastive patterns by preserving the local structures while disregarding the global structures present in normal images, (2) a self-supervised reweighting mechanism to handle the challenge of long-tailed and unlabeled contrastive patterns generated by PatchDiff, and (3) a lightweight patch-level detector to efficiently distinguish the normal patterns and reweighted contrastive patterns. The generation results of PatchDiff effectively expose various types of anomaly patterns, e.g. structural and logical anomaly patterns. In addition, extensive experiments on both MVTec AD and MVTec LOCO datasets also support the aforementioned observation and demonstrate that GRAD achieves competitive anomaly detection accuracy and superior inference speed.

Calibration estimation of distribution function based on multidimensional scaling of auxiliary information

The distribution function is a functional parameter of great interest in many research areas, such as medicine or economics. Among other properties, it facilitates the estimation of parameters such as quantiles. Accordingly, techniques are needed to estimate this function efficiently. Survey statisticians have access to large, high-dimension databases and use them to optimise the estimates obtained. One way to incorporate auxiliary information in the estimation stage is through the calibration method, which was initially designed to estimate totals and means and consists of adjusting new sample weights in order to reduce the variance of estimators. However, calibration techniques may be subject to over-calibration, i.e. the loss of efficiency when high-dimension auxiliary data sets are incorporated. Although alternative approaches have been proposed, in which the calibration method incorporates auxiliary information in the estimation of the distribution function, these alternatives do not seek to incorporate qualitative auxiliary information, which must be introduced in the usual way through dummy variables. However, this workaround can greatly increase the dimension of the auxiliary information, producing either over-calibration or even incompatible calibration constraints. In this article, we propose adapting the calibration method through multidimensional scaling, in order to incorporate quantitative and qualitative information, thus avoiding the negative consequences of over-calibration in the estimation of the distribution function.

Gradient-based domain-augmented meta-learning single-domain generalization for fault diagnosis under variable operating conditions

Equipment operating conditions, referred to as domains, can induce domain drift in monitoring data, affecting data-driven fault diagnosis. Researchers have explored multi-domain generalization methods to tackle this issue. However, in actual industrial scenarios, the availability of fault data may be limited to a specific condition due to the cost or feasibility constraints associated with collecting extensive monitoring data. This limitation hampers the generalization ability of these methods, posing a major challenge for robust fault diagnosis under variable operating conditions. To address this challenge, we proposed a gradient-based domain-augmented meta-learning (GDM) single-domain generalization method. We analyze the restrictions of generating fake domains and construct a domain-augmented loss by evaluating diagnostic tasks minimization, semantic consistency, and distribution diversity for fake samples. Using a gradient-based technique, fake domains are generated iteratively, providing diverse fault knowledge for improved generalization. Instead of using time-consuming ensemble methods, we develop a novel meta-learning method to train a highly efficient and generalizable model, relaxing the requirement for auxiliary datasets in existing meta-learning methods. Two case studies consistently demonstrate the effectiveness and superiority of the proposed GDM method. Our findings suggest that this study offers a promising and competitive solution for single-domain generalization in fault diagnosis within real industrial scenarios.

Deeper than DEEP: a spectroscopic survey of z &amp;gt; 3 Ly α emitters in the Extended Groth Strip

ABSTRACT We present a spectroscopic survey of Ly α emitters in the Extended Groth Strip (EGS) field, targeting the regime near the Epoch of Reionization. Using Keck/DEep Imaging Multi-Object Spectrograph, we observed 947 high-z candidates with photometric redshifts from 3 &amp;lt; zphot &amp;lt; 7 and down to an H-band (Hubble Space Telescope/Wide Field Camera 3 F160W) magnitude limit of &amp;lt;27.5. Observations were taken over the course of eight nights, with integration times ranging from 4 to 7.8 h. Our survey secured 137 unique redshifts, 126 of which are Ly α emitters at 2.8 &amp;lt; z &amp;lt; 6.3 with a mean redshift of $\overline{z} = 4.3$. We provide a comprehensive redshift catalogue for our targets, as well as the reduced one- and two-dimensional spectra for each object. These observations will provide an important auxiliary data set for the JWST Directors Discretionary Early Release Science programme the Cosmic Evolution Early Release Science Survey, which recently completed near- and mid-infrared imaging and spectroscopy of galaxies in the EGS field.