Background Model Research Articles

On Training Targets and Activation Functions for Deep Representation Learning in Text-Dependent Speaker Verification

Deep representation learning has gained significant momentum in advancing text-dependent speaker verification (TD-SV) systems. When designing deep neural networks (DNN) for extracting bottleneck (BN) features, the key considerations include training targets, activation functions, and loss functions. In this paper, we systematically study the impact of these choices on the performance of TD-SV. For training targets, we consider speaker identity, time-contrastive learning (TCL), and auto-regressive prediction coding, with the first being supervised and the last two being self-supervised. Furthermore, we study a range of loss functions when speaker identity is used as the training target. With regard to activation functions, we study the widely used sigmoid function, rectified linear unit (ReLU), and Gaussian error linear unit (GELU). We experimentally show that GELU is able to reduce the error rates of TD-SV significantly compared to sigmoid, irrespective of the training target. Among the three training targets, TCL performs the best. Among the various loss functions, cross-entropy, joint-softmax, and focal loss functions outperform the others. Finally, the score-level fusion of different systems is also able to reduce the error rates. To evaluate the representation learning methods, experiments are conducted on the RedDots 2016 challenge database consisting of short utterances for TD-SV systems based on classic Gaussian mixture model-universal background model (GMM-UBM) and i-vector methods.

A black hole toy model with non-local and boundary modes from non-trivial boundary conditions

We study gauge theories between two parallel boundaries with non-trivial boundary conditions, which serve as a toy model for black hole background with two boundaries near the horizon and infinite, aiming for a better understanding of the Bekenstein–Hawking entropy. The new set of boundary conditions allows boundary modes and non-local modes that interplay between the two boundaries. Those boundary modes and Wilson lines stretched between the two boundaries are carefully analyzed and are confirmed as physical variables in the phase space. Along with bulk fluctuation modes and topological modes, the partition function and entropy of all physical modes are evaluated via Euclidean path integral. It is shown that there are transitions between the dominance of different modes as we vary the temperature. The boundary fluctuation modes whose entropy is proportional to the volume dominate at high temperatures, and the boundary-area scaled boundary modes and Wilson lines are the more important at low temperatures. At super-low temperatures, when all the fluctuation modes die off, we see the topological modes whose entropy is the logarithm of the length scales of the system. The boundary modes and non-local modes should have their counterparts in a black hole system with similar boundary conditions, which might provide important hints for black hole physics.

Measuring the cosmic X-ray background accurately

Synthesis models of the diffuse Cosmic X-ray Background (CXB) suggest that it can be resolved into discrete sources, primarily Active Galactic Nuclei (AGNs). Measuring the CXB accurately offers a unique probe to study the AGN population in the nearby Universe. Current hard X-ray instruments suffer from the time-dependent background and cross-calibration issues. As a result, their measurements of the CXB normalization have an uncertainty of the order of sim 15%. In this paper, we present the concept and simulated performances of a CXB detector, which could be operated on different platforms. With a 16-Unit CubeSat mission running for more than two years in space, such a detector could measure the CXB normalization with sim 1% uncertainty.

Regional ionospheric TEC modeling during geomagnetic storm in August 2021- data fusion using multi-instrument observations

Regional ionospheric Total Electron Content (TEC) models are most significant to investigate the variability of ionosphere during all-weather conditions. Since the ground-based TEC measurements are not available in oceanic and polar regions, space-based observations are incorporated to increase the positional accuracy and reliability of Global Ionospheric TEC Maps (GIMs). Therefore, data fusion techniques are necessary to enhance the prediction capability of regional ionospheric models by considering multi-source data. In view of this, an attempt is made to investigate Equatorial Ionization Anomaly (EIA) structures in low-latitudinal Indian region using multi-instrument data. Data observations from background model CODEGIM, available International GNSS Service (IGS) stations in India, GNSS receivers at KLEF and Tripura stations, FORMOSAT/COSMIC radio occultation and SWARM during geomagnetic storm period August 26–28, 2021 are used for the study. Spherical Harmonics Function (SHF) of order 4 is employed along with Weighted Least Squares analysis (WLS), where separate weight constraint is computed for each data source. TEC maps of spatial resolution 5° x 5° and temporal resolution of 1 h are generated to illustrate the dynamic behavior of ionosphere. The standard deviation for 3 days is obtained as 0.42, 0.61 and 1.43 TECU respectively. TEC depletions and enhancements are observed on August 27 and 28, with mean values of 12.74 and 15.27 TECU respectively, demonstrating negative and positive storm effects.

Constraints on heavy dark matter annihilation and decay from electron and positron cosmic ray spectra

Annihilation or decay of dark matter (DM) could contribute to the electron and positron cosmic-ray flux, allowing for constraints on DM parameters from its measurement. CALET is directly measuring the energy spectrum of electron+positron cosmic rays up into the TeV region most important for studying heavy DM, while AMS-02 provides a positron-only spectrum below the TeV range. Limits on DM annihilation and decay well into the TeV mass range have been derived from a combined analysis of both data-sets with an astrophysical background model including pulsars as the origin of the positron excess and individual nearby supernova remnant sources.

The low-energy spectrum in DAMIC at SNOLAB

The DAMIC experiment employs large-area, thick charge-coupled devices (CCDs) to search for the interactions of low-mass dark matter particles in the galactic halo with silicon atoms in the CCD target. From 2017 to 2019, DAMIC collected data with a seven-CCD array (40-gram target) installed in the SNOLAB underground laboratory. We report dark-matter search results, including a conspicuous excess of events above the background model below 200V_{ee}Vee, whose origin remains unknown. We present details of the published spectral analysis, and update on the deployment of skipper CCDs to perform a more precise measurement by early 2023.

Search for low mass WIMP dark matter with DarkSide-50

DarkSide-50 is a direct dark matter detection experiment at Laboratori Nazionali del Gran Sasso that uses argon as the target material. Exploiting the ionization signal from a dual-phase time projection chamber filled with low-radioactivity argon, it has set the most stringent exclusion limit on WIMPs with a mass of few \mathrm{GeV}/c^2GeV/c2. A new analysis has recently been carried out with a larger exposure, profiting from an improved understanding of the detector response and background model. An improvement of about a factor 10 in sensitivity is expected for a WIMP mass of 2 GeV/c^22.

The structural tuning of the convolutional neural network forspeaker identification in mel frequency cepstrumcoefficients space

As a result of the literature analysis, the main methods for speaker identification from speech signals were defined. These are statistical methods based on Gaussian mixture model and a universal background model, as well as neural network methods, in particular, using convolutional or Siamese neural networks. The main characteristics of these methods are the recognition performance, a number of parameters, and the training time. High recognition performance is achieved by using convolutional neural networks, but a number of parameters of these networks are much higher than for statistical methods, although lower than for Siamese neural networks. A large number of parameters require a large training set, which is not always available for the researcher. In addition, despite the effectiveness of convolutional neural networks, model size and inference efficiency remain important fordevices with a limited source of computing power, such as peripheral or mobile devices. Therefore, the aspects of tuning of the structure of existing convolutional neural networks are relevant for research. In this work, we have performed a structural tuning of an existing convolutional neural network based on the VGGNet architecture for speaker identification in the space of mel frequency cepstrum coefficients. The aim of the work was to reduce the number of neural network parameters and, as a result, to reduce the network training time, provided that the recognition performance is sufficient (the correct recognition is above 95 %). The neural network proposed as a result of structural tuning has fewer layers than the architecture of the basic neural network. Insteadof the ReLU activation function, the related Leaky ReLU function with a parameter of 0.1 was used. The number of filters and the size of kernels in convolutional layers are changed. The size of kernels for the max pooling layer has been increased. It is proposedto use the averaging of the results of each convolution to input a two-dimensional convolution results to a fully connected layer with the Softmax activation function. The performed experiment showed that the number of parameters of the proposed neural network is 29 % less than the number of parameters of the basic neural network, provided that the speaker recognition performance is almost the same. In addition, the training time of the proposed and basic neural networks was evaluated on five datasets of audio recordings corresponding to different numbers of speakers. The training time of the proposed network was reduced by 10-39 % compared to the basic neural network. The results of the research show the advisability of the structural tuning of the convolutional neural network for devices with a limited source of computing, namely, peripheral or mobile devices.

The NANOGrav 15 yr Data Set: Bayesian Limits on Gravitational Waves from Individual Supermassive Black Hole Binaries

Evidence for a low-frequency stochastic gravitational-wave background has recently been reported based on analyses of pulsar timing array data. The most likely source of such a background is a population of supermassive black hole binaries, the loudest of which may be individually detected in these data sets. Here we present the search for individual supermassive black hole binaries in the NANOGrav 15 yr data set. We introduce several new techniques, which enhance the efficiency and modeling accuracy of the analysis. The search uncovered weak evidence for two candidate signals, one with a gravitational-wave frequency of ∼4 nHz, and another at ∼170 nHz. The significance of the low-frequency candidate was greatly diminished when Hellings–Downs correlations were included in the background model. The high-frequency candidate was discounted due to the lack of a plausible host galaxy, the unlikely astrophysical prior odds of finding such a source, and since most of its support comes from a single pulsar with a commensurate binary period. Finding no compelling evidence for signals from individual binary systems, we place upper limits on the strain amplitude of gravitational waves emitted by such systems. At our most sensitive frequency of 6 nHz, we place a sky-averaged 95% upper limit of 8 × 10−15 on the strain amplitude. We also calculate an exclusion volume and a corresponding effective radius, within which we can rule out the presence of black hole binaries emitting at a given frequency.

Unsupervised video rain streaks removal with deep foreground–background modeling

Outdoor video rain streaks removal is an important inverse problem in video processing that benefits subsequent applications. Traditional methods utilize prior information with interpretable domain knowledge while they are not tenable to capture complex structures of real-world videos. Deep learning methods learn a deraining mapping with a large model capacity brought by deep neural networks and their performances highly depend on the volume and diversity of training data. To address the challenging video deraining problem, we suggest an unsupervised video rain streaks removal method by solely using the observed rainy video. For the complex clean video, inspired by the classical foreground–background decomposition, we employ a deep convolutional neural network to capture the moving foreground and a disentangled deep spatial–temporal network with an affine operator to capture the underlying low-rank structure of the dynamic background. The foreground and background components are well balanced by a learnable probability mask. For structured rain streaks, we introduce a learnable total variation regularization whose parameters (i.e., rain directions) can be unsupervisedly learned. The deep modeling of the complex clean video and the simple yet effective modeling of structured rain streaks under the physical interpretable decomposition framework, which benefit each other in nature, are organically integrated to boost the deraining performance. Extensive experiments on synthetic and real-world rainy videos demonstrate the superiority of our method over state-of-the-art traditional and deep learning-based video deraining methods.

Ionospheric Vertical Correlation Distance Calculation Based on COSMIC Electron Density Profile Data

AbstractBackground error covariance is an important part of ionospheric assimilation and contains correlation information of the ionospheric background field and error information of the model background field, determines the influence degree of observations on models, and controls the vertical electron density distribution of assimilation results. For more accurate background error covariance determination along the ionosphere vertical direction, we used Constellation Observing System for Meteorology Ionosphere and Climate occultation electron density profile data for vertical correlation distance calculation considering IRI 2016 model errors. The vertical direction correlations are asymmetrical above and below the reference altitude, and with increasing altitude, the vertical correlation distance growth rate may decline or remain unchanged. There are correlation differences in different solar activity years, varying with local time, geomagnetic latitude, and altitude. The vertical correlation distances in high‐solar activity years are greater than that in low‐solar activity years. The correlation distances difference at different times decreases with latitude, with the largest correlation distances at low latitudes. The nighttime distances vary slightly with latitude, while the sunrise correlation distances vary the most relative to other periods. At the same altitude, the correlation distances in the daytime low‐latitude region are the smallest overall, while the correlation distances between the daytime mid‐latitude region and sunrise time are the largest. The point where vertical correlation distances growth stabilize may be correlated with ionospheric hmF2 and upper transition. The IRI model is the most widely used background model in ionospheric assimilation, and the study results could facilitate more accurate vertical background error covariance for ionospheric assimilation.

A New Surveillance and Security Alert System Based on Real-Time Motion Detection

Due to the frequent and widespread instances of burglary, effective and accurate intrusion detection with an alarm system is now necessary. In this study, the proposed system is used for secret and vital places that need a good monitoring system based on human motion detection. The system can protect museums that contain critical historical objects and libraries with old historical books. The security alert system is also essential to the protection of banks. The system uses a two-frame difference algorithm and a simple background modeling algorithm for motion detection with the help of AForge.NET. Whenever the system detects motion, it interprets the frames of motion to determine humans by using YOLO. The security alert system activates the alerts whenever it detects human motion, like sending an email to an authorized person, triggering the LED alert and sound alert, recording a video of that place, taking a motion picture, and saving it within the databases. The system has the advantages of low cost, a user-friendly interface, and high security. The system was thoroughly tested to detect human movement in unauthorized areas. The system successfully detected human activity indoors and outdoors and activated the alarms. Also, the system could detect human movement in all four directions and alert observers directly at the time of the incident. In low light or minimum illumination, the system can determine human motion. The evaluation results showed that the system's accuracy in detecting human movement was 95%, which shows that the system is able to protect vital and secret places and can successfully monitor areas.

Point-of-care testing for adult out-of-hospital cardiac arrest resuscitated at the ED to predict ROSC: Development and external validation of POC-ED-ROSC model

Background and importanceMost prediction models, like return of spontaneous circulation (ROSC) after cardiac arrest (RACA) or Utstein-based (UB)-ROSC score, were developed for prehospital settings to predict the probability of ROSC in patients with out-of-hospital cardiac arrest (OHCA). A prediction model has been lacking for the probability of ROSC in patients with OHCA at emergency departments (EDs). ObjectiveIn the present study, a point-of-care (POC) testing-based model, POC-ED-ROSC, was developed and validated for predicting ROSC of OHCA at EDs. Design, settings and participantsProspectively collected data for adult OHCA patients between 2015 and 2020 were analysed. POC blood gas analysis obtained within 5 min of ED arrival was used. Outcomes measure and analysisThe primary outcome was ROSC. In the derivation cohort, multivariable logistic regression was used to develop the POC-ED-ROSC model. In the temporally split validation cohort, the discriminative performance of the POC-ED-ROSC model was assessed using the area under the receiver operating characteristic (ROC) curve (AUC) and compared with RACA or UB-ROSC score using DeLong test. Main resultsThe study included 606 and 270 patients in the derivation and validation cohorts, respectively. In the total cohort, 471 patients achieved ROSC. Age, initial cardiac rhythm at ED, pre-hospital resuscitation duration, and POC testing-measured blood levels of lactate, potassium and glucose were significant predictors included in the POC-ED-ROSC model. The model was validated with fair discriminative performance (AUC: 0.75, 95% confidence interval [CI]: 0.69–0.81) with no significant differences from RACA (AUC: 0.68, 95% CI: 0.62–0.74) or UB-ROSC score (AUC: 0.74, 95% CI: 0.68–0.79). ConclusionUsing only six easily accessible variables, the POC-ED-ROSC model can predict ROSC for OHCA resuscitated at ED with fair accuracy.

Deep background-mismodeling-learned reconstruction for high-accuracy fluorescence diffuse optical tomography.

We present a deep background-mismodeling-learned reconstruction framework for high-accuracy fluorescence diffuse optical tomography (FDOT). A learnable regularizer incorporating background mismodeling is formulated in the form of certain mathematical constraints. The regularizer is then learned to obtain the background mismodeling automatically using a physics-informed deep network implicitly. Here, a deep-unrolled FIST-Net for optimizing L1-FDOT is specially designed to obtain fewer learning parameters. Experiments show that the accuracy of FDOT is significantly improved via implicitly learning the background mismodeling, which proves the validity of the deep background-mismodeling-learned reconstruction. The proposed framework can also be used as a general method to improve a class of image modalities based on linear inverse problems with unknown background modeling errors.

Assimilation and Inversion of Ionospheric Electron Density Data Using Lightning Whistlers

The data assimilation algorithm is a common algorithm in space weather research. In this paper, the time-frequency information in the dispersion spectrum of lightning whistlers received by the ZH-1 satellite is used as the observed value, and the international reference ionospheric model serves as the background model to construct the calculation model of the propagation time of lightning whistlers in the ionosphere. Kalman filtering is adopted to assimilate the electron density distribution along the propagation path of lightning whistlers. The results show that the situation where the electron density of the background model deviates greatly from the true value is significantly improved through data assimilation. The electron density after assimilation is in good agreement with the true value, which effectively helps realize the process of using observed values to correct the background value. On this basis, the influence of the frequency difference on the assimilation inversion effect is studied, and the results show that the assimilation effect is worse when the frequency difference between frequency points is less than 1 kHz.

Anomaly Detection of Remote Sensing Images Based on the Channel Attention Mechanism and LRX

Anomaly detection of remote sensing images has gained significant attention in remote sensing image processing due to their rich spectral information. The Local RX (LRX) algorithm, derived from the Reed–Xiaoli (RX) algorithm, is a hyperspectral anomaly detection method that focuses on identifying anomalous pixels in hyperspectral images by exploiting local statistics and background modeling. However, it is still susceptible to the noises in the Hyperspectral Images (HSIs), which limits its detection performance. To address this problem, a hyperspectral anomaly detection algorithm based on channel attention mechanism and LRX is proposed in this paper. The HSI is feed into the auto-encoder network that is constrained by the channel attention module to generate a more representative reconstructed image that better captures the characteristics of different land covers and has less noises. The channel attention module in the auto-encoder network aims to explore the effective spectral bands corresponding to different land covers. Subsequently, the LRX algorithm is utilized for anomaly detection on the reconstructed image obtained from the auto-encoder network with the channel attention mechanism, which avoids the influence of noises on the anomaly detection results and improves the anomaly detection performance. The experiments are conducted on three HSIs to verify the performance of the proposed method. The proposed hyperspectral anomaly detection method achieves higher Area Under Curve (AUC) values of 0.9871, 0.9916 and 0.9642 on HYDICE urban dataset, AVIRIS aircraft dataset and Salinas Valley dataset, respectively, compared with other six methods. The experimental results demonstrate that the proposed algorithm has better anomaly detection performance than LRX and other algorithms.

Energy Efficiency Model Construction of Building Carbon Neutrality Design

We aim to create a feasible quantitative method to calculate the energy efficiency of building designs that are carbon-neutral and to develop a workable way of calculating energy efficiency in buildings that achieve carbon neutrality and the system for such a building’s design energy efficiency function. This paper first clarifies the idea of the design energy efficiency function for a carbon-neutral building over its whole life cycle. Subsequently, through the efficient analysis of carbon-neutral design dimension measures, this paper summarizes and integrates the mature theories of various disciplines, puts forward the energy efficiency function model of carbon-neutral design background, propulsion, and coverage, and implements the energy efficiency function model of carbon-neutral design in the whole life cycle of buildings. The index value of a building’s carbon emission factor is established based on the carbon accounting factor published by the Intergovernmental Panel on Climate Change, and a carbon neutrality energy efficiency model for buildings over the duration of their whole life cycle is constructed. The results were as follows. 1. Technology energy efficiency is far better than scale energy efficiency and comprehensive energy efficiency. 2. The better the energy efficiency value inside the building stage, the less consumption and the higher the production. 3. Construction is when technical energy is used the least. This paper refers to a systematic design method that makes the level of building carbon neutrality design technologically advanced with the aid of all types of big data related to the building life cycle and various innovative design theories in order to fully represent the fundamental level, development potential, and the effectiveness of choosing the strategy of building carbon neutrality.

A novel rat model of Dravet syndrome recapitulates clinical hallmarks

Dravet syndrome (DS) is a debilitating infantile epileptic encephalopathy characterized by seizures induced by high body temperature (hyperthermia), sudden unexpected death in epilepsy (SUDEP), cognitive impairment, and behavioral disturbances. The most common cause of DS is haploinsufficiency of the SCN1A gene, which encodes the voltage-gated sodium channel Nav1.1. In current mouse models of DS, the epileptic phenotype is strictly dependent on the genetic background and most mouse models exhibit drastically higher SUDEP rates than patients. Therefore, we sought to develop an alternative animal model for DS. Here, we report the generation and characterization of a Scn1a halploinsufficiency rat model of DS by disrupting the Scn1a allele. Scn1a+/− rats show reduced Scn1a expression in the cerebral cortex, hippocampus and thalamus. Homozygous null rats die prematurely. Heterozygous animals are highly susceptible to heat-induced seizures, the clinical hallmark of DS, but are otherwise normal in survival, growth, and behavior without seizure induction. Hyperthermia-induced seizures activate distinct sets of neurons in the hippocampus and hypothalamus in Scn1a+/− rats. Electroencephalogram (EEG) recordings in Scn1a+/− rats reveal characteristic ictal EEG with high amplitude bursts with significantly increased delta and theta power. After the initial hyperthermia-induced seizures, non-convulsive, and convulsive seizures occur spontaneously in Scn1a+/− rats. In conclusion, we generate a Scn1a haploinsufficiency rat model with phenotypes closely resembling DS, providing a unique platform for establishing therapies for DS.

A decoding-based method for fast background filtering of roadside LiDAR data

This study is part of an investigation on application of LiDAR sensors to roadside to create a human-in-the-loop system that actively protects all road users, especially pedestrians and cyclists. The primary challenge for implementing such a system in the field is its computational efficiency. Unlike on-board LiDAR sensors used in autonomous vehicles, roadside applications must perform complete background filtering and monitor entire real-time traffic movements. This paper presents an innovative method for fast filtering background objects from roadside LiDAR data and demonstrates its capabilities to achieve higher accuracy, faster processing speed, and require less data storage. The proposed method is innovative in that it embeds background filtering within the decoding process to exclude irrelevant information, thus saving computational expense on analyzing background points that are not of interest. A 2D channel-azimuth background table is generated by learning the critical distance information of both static and dynamic backgrounds, which is more accurate than traditional background models that only focus on static backgrounds. Performance measurements and comparison analysis were conducted using data from our testbeds and open source. Compared with the state-of-the-art methods, the new method achieves faster processing speed (0.65 ms/frame and 0.90 ms/frame) and requires less data storage (225 KB and 450 KB memory) when processing 16- and 32-laser LiDAR data collected at 10 Hz.

Attraction Recommendation Based on Tourism Context Modeling and Multi‐neural Collaborative Filtering Algorithm

The key to generating tailored suggestions in the scenic location recommendation scene is how to model the diverse tourism environments in order to correctly acquire visitor preferences and scenic spot tourism features. However, most existing recommendation algorithms focus on the spatiotemporal background modeling of historical tourism trajectories. Only tourists' preferences are understood, and the rich heterogeneous tourism information such as tourists' personal tourism constraints and scenic spots' tourism attributes are ignored. In this paper, we proposed a multiple neural collaborative filtering attraction recommendation architecture (MNCF‐AR). To begin, we learn the tourism feature representation of tourists by modeling diverse tourism contexts, and then we create the tourism trajectory background of tourists using a large number of actual tourism logs to achieve the entire feature representation of tourists. Second, to learn the feature vector of scenic spots in the context, the tourism heterogeneous network map is used to build the scenic spot attribute background, the self‐attention network to learn the scenic spot sequence learning, and a neural network to project each scenic spot into a unified potential feature space. Finally, the multi‐neural collaborative filtering method is utilized to forecast the difference in scores between visitors and scenic sites so that tailored scenic spots may be recommended. Extensive experiments on mainstream datasets, such as MaFengwo, New York, Tokyo and Xi'an, show that the proposed method can effectively and accurately recommend attractions for users. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.