Data Distribution Research Articles

Photobiomodulation with laser and led on mesenchymal stem cells viability and wound closure in vitro.

Mesenchymal stem cells can differentiate into specific cell lineages in the tissue repair process. Photobiomodulation with laser and LED is used to treat several comorbidities, can interfere in cell proliferation and viability, in addition to promoting responses related to the physical parameters adopted. Evaluate and compare the effects of laser and LED on mesenchymal cells, with different energy doses and different wavelengths, in addition to viability and wound closure. Mesenchymal stem cells derived from human adipocytes were irradiated with laser (energy of 0.5J, 2J and 4J, wavelength of 660nm and 830nm), and LED (energy of 0.5J, 2J and 4J, where lengths are 630nm and 850nm). The wound closure process was evaluated through monitoring the reduction of the lesion area in vitro. Viability was determined by analysis with Hoechst and Propidium Iodide markers, and quantification of viable and non-viable cells respectively Data distributions were analyzed using the Shapiro-Wilk test. Homogeneity was analyzed using Levene's test. The comparison between the parameters used was analyzed using the Two-way ANOVA test. The T test was applied to data relating to viability and lesion area. For LED photobiomodulation, only the 630nm wavelength obtained a significant result in 24, 48 and 72h (p = 0,027; p = 0,024; p = 0,009). The results related to the in vitro wound closure test indicate that both photobiomodulation with laser and LED demonstrated significant results considering the time it takes to approach the edges (p < 0.05). Considering the in vitro experimental conditions of the study, it is possible to conclude that the physical parameters of photobiomodulation, such as energy and wavelength, with laser or LED in mesenchymal stem cells, can play a potential role in cell viability and wound closure.

Statistical analysis for chemical compound based on several species of aquilaria essential oil

The paper examines the characterization of Aquilaria essential oils from different species, namely Aquilaria malaccensis, Aquilaria beccariana, Aquilaria crassna, and Aquilaria subintegra, renowned for agarwood production in Malaysia. Gas chromatography-mass spectrometry (GC-MS) and gas chromatography-flame ionization detector (GC-FID) were employed for extracting essential oil data, facilitating compound identification. Subsequently, a preliminary analysis focused on classifying significant chemical compounds in the samples. The study then utilized boxplot pre-processing for visualizing and interpreting data distribution. The statistical analyses were performed using MATLAB software version R2021b, considering two key parameters which are the peak area (%) of significant chemical compounds and the classification of Aquilaria species (A. beccariana, A. malaccensis, A. crassna, and A. subintegra) based on their chemical composition. The results, presented through boxplot analyses, demonstrated a clear representation of the parameters and their distribution in the data. This method not only confirmed the potential of boxplot analysis in statistical evaluation of significant compounds in Aquilaria essential oil but also suggested its applicability for further classification work.

A deep learning approach based on domain generalization for mooring tension prediction in floating structures

Generalization remains a significant challenge in the application of deep learning methods, particularly in predicting mooring line tension for moored floating structures. The dynamic and complex nature of marine environments makes accurate prediction of mooring line tension a formidable task. Existing deep learning models, constrained by the data distribution of their training sets, typically perform well within the scope of these datasets. However, their limited generalization capability becomes apparent when confronted with unknown and changing marine conditions, resulting in a decline in prediction accuracy. To address this issue, this paper proposes a Domain Generalization (DG) model approach, which incorporates a distribution matching regularization term to the model's loss function, aimed at reducing distributional disparities between data from different domains. The proposed method was compared with existing Deep Neural Network (DNN) model method through validation with engineering case studies. The results indicate that the DG model method significantly surpasses the DNN model in terms of generalization capabilities. The mooring line tension prediction model developed herein holds potential for constructing digital twins, offering near real-time data support for moored floating engineering projects. This facilitates timely and accurate decision-making in response to environmental changes.

Low-, moderate-, and high-risk obesity in association with cost drivers, costs over the lifecycle, and life expectancy

BackgroundThe prevalence of low-, moderate-, and high-risk obesity has been increasing globally. Our aim was to estimate the societal burden of these three obesity classes in the Austrian population by taking a societal–annual perspective and an individual–life-cycle perspective. Secondly, we sought to identify the respective cost drivers and the effects on life expectancy.MethodsWe used population-weighted survey data on the distribution of body mass index (BMI) and data on relative risks regarding 83 diseases. Using fractional polynomial regressions, we estimated relative risks per BMI unit for about 30 cost-intensive diseases up to BMI values of 50. The approach for the cost analysis was based on the use of population-attributable fractions applied to direct medical and indirect cost data. Macro-disease-specific data regarding cost factors came from cost-of-illness statistics and administrative sources.ResultsAbout 8.2% of deaths and 4.6% of health expenditure are attributable to obesity in Austria in 2019, causing 0.61% of GDP loss. A third of annual direct and indirect costs came from class 2 and class 3 obesity. From an individual perspective, life-cycle costs of class 2 and class 3 obesity were 1.9 and 3.6 times the costs of class 1 obesity, respectively. At the age of 45, people with high-risk obesity are about to lose approximately 4.9 life years and 9.7 quality-adjusted life years.ConclusionsWe have extended the method of population-attributable fractions, allowing us to estimate differences in the life-cycle direct medical and indirect costs between low-, moderate-, and high-risk obesity. We found that the lifetime costs of obesity are strikingly different between obesity classes, which has been veiled in the societal—annual perspective. Our findings provide a foundation for evaluating public health interventions aimed at combating obesity in order to mitigate the escalating individual and societal burden resulting from obesity prevalence and complications in the future.

A metaheuristic algorithmic framework for solving the hybrid flow shop scheduling problem with unrelated parallel machines

The hybrid flow shop scheduling problem (HFSP), as a realistic extension of the classical flow shop scheduling problem, widely exists in real-world industrial production systems. In practice, the fact that machines are unrelated is important and cannot be neglected. This study focuses on the HFSP with unrelated parallel machines (HFSP-UPM) to minimize the makespan. To address this problem, a hybrid representation that combines single-sequence coding and full-sequence coding is developed to search for solution space that is not covered by common encoding methods. An initialization block integrating random heuristic strategies is proposed for improving the quality of the initial sparrow swarm. To improve the diversity of a sparrow swarm further, a perturbation block embedded with a set of historical best positions and an enhancement strategy are developed. A critical set based local search block is designed for high-intensity local exploitation of promising regions when necessary. Several benchmark cases from the literature as well as some randomly generated instances characterized by the distribution of real data from factory studies are employed to participate in the test. The test results reveal the effectiveness of the proposed perturbation block and local search block. Compared to state-of-the-art metaheuristic algorithms, the enhanced sparrow search algorithm (ESSA) demonstrates higher convergence accuracy when handling instances. The value of the gap between a solution found by the ESSA and a feasible solution found by the mixed-integer programming (MIP) model can reach 0.6%.

Secure Processing and Distribution of Data Managed on Private InterPlanetary File System Using Zero-Knowledge Proofs

Secure Processing and Distribution of Data Managed on Private InterPlanetary File System Using Zero-Knowledge Proofs

Physics-informed neural network for turbulent flow reconstruction in composite porous-fluid systems

This study explores the implementation of physics-informed neural networks (PINNs) to analyze turbulent flow in composite porous-fluid systems. These systems are composed of a fluid-saturated porous medium and an adjacent fluid, where the flow properties are exchanged across the porous-fluid interface. The segregated PINN model employs a novel approach combining supervised learning and enforces fidelity to flow physics through penalization by the Reynolds-averaged Navier-Stokes (RANS) equations. Two cases were simulated for this purpose: solid block, i.e. porous media with zero porosity, and porous block with a defined porosity. The effect of providing internal training data on the accuracy of the PINN predictions for prominent flow features, including flow leakage, channeling effect and wake recirculation was investigated. Additionally, L2 norm error, which evaluates the prediction accuracy for flow variables was studied. Furthermore, PINN training time in both cases with internal training data was considered in this study. Results showed that the PINN model predictions with second-order internal training data achieved high accuracy for the prominent flow features compared to the RANS data, within a 20% L2 norm error of second-order statistics in the solid block case. In addition, for the porous block case, providing training data at the porous-fluid interface showed errors of 18.04% and 19.94% for second-order statistics, representing an increase in prediction accuracy by 7% compared to without interface training data. The study elucidates the impact of the internal training data distribution on the PINN training in complex turbulent flow dynamics, underscoring the necessity of turbulent second-order statistics variables in PINN training and an additional velocity gradient treatment to enhance PINN prediction.

Expowo: An R package for mining global plant diversity and distribution data

AbstractPremiseData on plant distribution and diversity from natural history collections and taxonomic databases are increasingly becoming available online as exemplified by the Royal Botanic Gardens, Kew's Plants of the World Online (POWO) database. This growing accumulation of biodiversity information requires an advance in bioinformatic tools for accessing and processing the massive data for use in downstream science. We present herein expowo, an open‐source package that facilitates extracting and using botanical data from POWO.Methods and ResultsThe expowo package is implemented in R and designed to handle the entire vascular plant tree of life. It includes functions to readily distill taxonomic and distributional information about all families, genera, or species of vascular plants. It outputs a complete list of species in each genus of any plant family, with the associated original publication, synonyms, and distribution, and plots global maps of species richness at the country and botanical country levels, as well as graphs displaying species‐discovery accumulation curves and nomenclatural changes over time. To demonstrate expowo's strengths in producing easy‐to‐handle outputs, we also show empirical examples from a set of biodiverse countries and representative species‐rich and ecologically important angiosperm families.ConclusionsBy harnessing bioinformatic tools that accommodate varying levels of R programming proficiency, expowo functions assist users who have limited R programming expertise in efficiently distilling specific botanical information from online sources and producing maps and graphics for the further interpretation of biogeographic and taxonomic patterns.

Toward compound fault diagnosis via EMAGAN and large kernel augmented few-shot learning

Bearings are essential in machinery. Damage to them can cause financial losses and safety risks at industrial sites. Therefore, it is necessary to design an accurate diagnostic model. Although many bearing fault diagnosis methods have been proposed recently, they still cannot meet the requirements of high-accurate prediction of bearing faults. There are several challenges in this: 1) In practical settings, gathering sufficient and balanced sample data for training diagnostic network models proves challenging. 2) The damage to bearings in real industrial production sites is not singular, and compound faults are also a huge challenge for diagnostic networks. To address these issues, this study introduces a novel fault diagnosis model called EMALKNet that integrates DCGAN with Efficient Multi-Scale Attention (EMAGAN) and RepLKNet-XL, enhancing the detection and analysis of bearing faults in industrial machinery. This model employs EMAGAN to explore the underlying distribution of raw data, thereby enlarging the fault sample pool and enhancing the model’s diagnostic capabilities; The large kernel structure of RepLKNet-XL is different from the current mainstream small kernel and has stronger representation extraction ability. The proposed method has been validated on the Paderborn University dataset and the Huazhong University of Science and Technology dataset.

A method for estimating the rotation errors of BDS broadcast ephemeris using global satellite laser ranging data

Abstract The BeiDou-3 Navigation Satellite System (BDS-3) generates navigation messages from a regional ground tracking network and inter-satellite links (ISL), relying on predicted station coordinates and Earth rotation parameters (ERPs). These procedures, however, may lead to orientation errors in the broadcast ephemeris, termed constellation rotation errors.&amp;#xD;Traditionally, these errors are monitored through Helmert transformation analysis comparing broadcast and post-processed orbits, a method that highlights discrepancies between the&amp;#xD;GNSS Terrestrial Reference Frame (TRF) and the International Terrestrial Reference Frame&amp;#xD;(ITRF) but suffers from a two-week evaluation delay. This study proposes a near real-time approach for estimating constellation rotation errors using high precision Satellite Laser-Ranging (SLR) data, representing the difference between GNSS TRF and ITRF through the&amp;#xD;International Laser Ranging Service (ILRS) TRF. The impact of SLR data quantity and distribution on rotation estimation accuracy is investigated, and the effectiveness of this method on BDS-3 orbits with ERP-induced orientation deficiencies is validated. Results show that the root mean square (RMS) of broadcast orbit errors and the orbit user ranging error&amp;#xD;(URE) reduce from 0.8/0.7/0.2 m to 0.1/0.1/0.1 m, respectively. This demonstrates the method’s capability for real-time monitoring and correcting orientation errors within BDS-3 navigation messages.&amp;#xD;

The association between fat distribution and α1-acid glycoprotein levels among adult females in the United States

BackgroundVisceral fat accumulation and obesity-induced chronic inflammation have been proposed as early markers for multiple disease states, especially in women. Nevertheless, the potential impact of fat distribution on α1-acid glycoprotein(AGP), a marker of inflammation, remains unclear. This research was conducted to investigate the relationships among obesity, fat distribution, and AGP levels.MethodsA cross-sectional observational study was performed using blood samples from adult females recruited through the National Health and Nutrition Examination Survey from 2015 to 2018. Serum levels of AGP were measured using the Tina-quant α-1-Acid Glycoprotein Gen.2 assay. Based on the fat distribution data obtained from dual-energy X-ray absorptiometry assessments, body mass index (BMI), total percent fat (TPF), android percent fat (APF), gynoid percent fat (GPF), android fat/gynoid fat ratio (AGR), visceral percent fat (VPF), subcutaneous percent fat (SPF), visceral fat/subcutaneous fat ratio (VSR) were used as dependent variables. To investigate the link between fat distribution and AGP, multivariate linear regression analysis was utilized. Furthermore, a sensitivity analysis was also performed.ResultsThe present study included 2,295 participants. After adjusting for covariates, BMI, TPF, APF, GPF, VPF, and SPF were found to be positively correlated with AGP levels (BMI: β = 23.65 95%CI:20.90–26.40; TPF: β = 25.91 95%CI:23.02–28.80; APF: β = 25.21 95%CI:22.49–27.93; GPF: β = 19.65 95%CI:16.96–22.34; VPF: β = 12.49 95%CI:9.08–15.90; SPF: β = 5.69, 95%CI:2.89–8.49; AGR: β = 21.14 95%CI:18.16–24.12; VSR: β = 9.35 95%CI:6.11–12.59, all P < 0.0001). All the above indicators exhibited a positive dose–response relationship with AGP. In terms of fat distribution, both AGR and VSR showed positive associations with AGP (P for trend < 0.0001). In particular, when compared to individuals in tertile 1 of AGR, participants in tertiles 2 and 3 had 13.42 mg/dL (95% CI 10.66–16.18) and 21.14 mg/dL (95% CI 18.16–24.12) higher AGP levels, respectively. Participants in the highest tertile of VSR were more likely to exhibit a 9.35 mg/dL increase in AGP compared to those in the lowest tertile (95% CI 6.11–12.59).ConclusionsOverall, this study revealed a positive dose-dependent relationship between fat proportion/distribution and AGP levels in women. These findings suggest that physicians can associate abnormal serum AGP and obesity with allow timely interventions.

A value assessment system for the conservation and utilization of decentralized distilleries in the architectural heritage of the brewing industry: a case study of Yibin city, China

ABSTRACT With a brewing history spanning over four millennia, Yibin city is adorned with myriad traditional distilleries that have dotted its landscape since ancient times. In recent years, industrial architectural heritage has approached the forefront of cultural heritage research, presenting new opportunities and challenges for the preservation and utilization of industrial architectural heritage sites, including distilleries. Employing field investigations and a multidisciplinary approach, this study uses data on 62 distilleries that are industrial architectural heritage sites. The analysis delves into the distribution, composition, and architectural features of the sample data, considering both production and nonproduction elements. Subsequently, the quantification of historical, societal, technological, artistic, and economic value enables an assessment, establishing an appropriate evaluation system. The research outcomes are intended to offer technical support for the conservation and utilization of heritage sites in the brewing industry, while concurrently serving as a reference for the preservation and utilization of distinctive architectural buildings in local industrial heritage sites.

A method for assessing the operational status of a microseismic monitoring system using energy distribution changes of observed data

Abstract The operational status of geophones plays a pivotal role in ensuring the accuracy and reliability of microseismic monitoring systems. However, conventional techniques used to evaluate the operational status of geophones require human intervention or significant time delays. To address this issue, we propose a method for online monitoring of geophone status using observed data obtained from a microseismic system. Firstly, the energy features of the preprocessed observation data are extracted via wavelet packet decomposition. Subsequently, the distribution parameters of energy features are obtained through loglogistic distribution fitting. These parameters are then applied to a change-point detection model, enabling the online monitoring of seismic geophones. In addition, we selected long short-term memory (LSTM) network to classify the operational status of the geophones, which is trained using the obtained energy distribution data and the time-frequency characteristics of the observed data. The experimental results indicate that the model achieves an accuracy of 98.33%, surpassing the 89.58% accuracy of the support vector machine (SVM). The proposed method not only contributes to online monitoring and precise determination of the operating status of detectors, but also has enormous application potential in other fields that require monitoring and evaluating the operating status of instruments.

The Spatiotemporal Correlation between Human Activity Intensity and the Evolution of Ecosystem Service Value in the Songnen Plain, China

For the main grain-producing areas worldwide that balance multi-tasks of grain production, ecological protection, and economic development, quantitatively revealing the correlation between human activity intensity (HAI) and ecosystem service value (ESV) is conducive to formulating adapted ecological protection policies and promoting the coordinated development of the regional economy, society, and ecosystem. In this paper, we took the Songnen Plain of China as an example, employed a modified Equivalent Factor Method (integrating socio-economic data, the normalized difference vegetation index (NDVI), and land use data), and the HAI Assessment Model (based on the data of land use, night-time light, and population spatial distribution) to measure the ESV and HAI in the Songnen Plain of China for the years 1990, 1995, 2000, 2005, 2010, 2015, and 2020. We further applied the standard deviational ellipse method, the coupled coordination degree model, and the bivariate spatial autocorrelation models to reveal the spatiotemporal dynamics and correlation characteristics of ESV and HAI. The results showed the following: (1) Temporally, the ESV declined from 950.96 billion yuan in 1990 to 836.31 billion yuan in 2015, and then increased to 864.60 billion yuan in 2020, with the total loss attributed to the significant decline in the ESV of the natural ecosystem. Spatially, the ESV in the western and northeastern regions was relatively high, with a significant increase in the northeast. (2) HAI showed an upward trend from 1990 to 2020, while the high HAI levels gradually shrank after reaching the peak in 2000. Low HAI levels were mainly distributed in the northeast and southwest, aligning with the ecological space, while high HAI levels were distributed in the middle and southeast. (3) The interaction between ESV and HAI was marked by a negative correlation, transitioning from conflict to coordination. The spatial pattern of HAI and ESV showed H (HAI)-L (ESV) and L-H clustering, with H-H and L-L scattered distributions. This study contributes to providing a framework, methods, and suggestions for the sustainable planning and utilization of land and ecological protection in order to offer scientific references for the Songnen Plain, other major grain-producing areas, and related studies.

EfficientNet-ECA: A lightweight network based on efficient channel attention for class-imbalanced welding defects classification

Welding defects recognition is crucial for ensuring weldment quality in robot arc welding. Nevertheless, due to unbalanced data distributions, limited computing resources in factory, as well as intraclass variability and interclass similarity among different welding defects, it is difficult to extract the most discriminative defect features from welding molten pool images on site, resulting in weak class-imbalanced defect recognition performance. To address the above issue, a novel lightweight network named EfficientNet-ECA is proposed for recognizing and classifying welding defects according to molten pool images of robot arc welding. Firstly, the EfficientNet-ECA network based on Efficient Channel Attention (ECA) is designed to extract the most discriminative features of different defects from molten pool images, where ECA is employed to enhance cross-channel information interactions. Secondly, a dynamic equalized focal loss function is proposed to both rebalance the loss contribution of class-imbalanced data of different defects and improve defects recognition accuracy. Subsequently, a class-imbalanced dataset containing eight typical welding defects is constructed to evaluate the EfficientNet-ECA model. Finally, the proposed model is comprehensively analyzed through ablation studies and compared with the existing state-of-the-art lightweight models, and results show that the proposed method exhibits better effectiveness and generalizability in classifying class-imbalanced defects across different welding scenarios, achieving the highest accuracy of 95.84% on a self-constructed AL5083 dataset and 96.50% on a publicly available SS304 dataset. Moreover, with fewer parameters and less complexity, the proposed method is more suitable for online welding quality monitoring in factory.

STAGE: a spatiotemporal-knowledge enhanced multi-task generative adversarial network (GAN) for trajectory generation

Individual trajectory data play a pivotal role in various application fields, such as urban planning, traffic control, and epidemic simulation. Despite the diverse means for data collection in current times, the real-world trajectory data in practical application remains severely limited due to concerns over personal privacy. In this study, we designed a Spatiotemporal-knowledge enhanced multi-TAsk GEnerative adversarial network (GAN), named STAGE, to generate synthetic trajectories that statistically resemble the real data without recycling personal information. In STAGE, we designed a multi-task generator with three stages of spatio-temporal generation tasks, i.e. activity-sequence generation task, township-level trajectory generation task, and neighborhood-level trajectory generation task, with the last one as the main task while the other two as auxiliary tasks. Meanwhile, we designed a spatial consistency loss in the adversarial training process to assess the spatial consistency of generated trajectories at different spatial scales. Experiment results show that compared to the baselines, trajectories generated by our method have closer data distributions to the real ones. We argued that the designs of spatiotemporal-knowledge enhanced generation tasks and training loss benefit the spatiotemporal generation processes, which help reproduce the temporal patterns of human daily activities and spatial distribution of human movements.

Bootstrap Method as a Tool for Analyzing Data with Atypical Distributions Deviating from Parametric Assumptions: Critique and Effectiveness Evaluation

In today’s research environment characterized by exponential data growth and increasing complexity, the selection of appropriate statistical tests, tailored to research objectives and data distributions, is paramount for rigorous analysis and accurate interpretation. This article explores the growing prominence of bootstrapping, an advanced statistical technique for multiple comparisons analysis, offering flexibility and customization by estimating sample distributions without assuming population distributions, thus serving as a valuable alternative to traditional methods in various data scenarios. Computer simulations were conducted using data from cardiovascular disease patients. Two approaches, spontaneous partly controlled simulation and fully constrained simulation using self-written R scripts, were utilized to generate datasets with specified distributions and analyze the data using tests for comparing more than two groups. The utilization of the bootstrap method greatly improves statistical analysis, especially in overcoming the constraints of conventional parametric tests. Our research showcased its effectiveness in comparing multiple scenarios, yielding strong findings across diverse distributions, even with minor inflation in p values. Serving as a valuable substitute for parametric approaches, bootstrap promotes careful consideration when rejecting hypotheses, thus fostering a deeper understanding of statistical nuances and bolstering analytical rigor.

Causality-inspired Domain Expansion network for single domain generalization

Most existing single domain generalization (SDG) methods focus on extending the distribution of the training data to achieve good generalizability in the target domain by synthesizing out-of-domain data. However, in practice, the generated data are insufficient to cover all possible shifts. When the target distributions are not covered, the spurious associations between irrelevant features and labels captured by a model will deteriorate the generalization performance. To alleviate this problem, in this paper, we first analyze that augmenting the source domain with samples of new styles (e.g., different backgrounds) can mitigate spurious associations between irrelevant features and labels from a causal perspective, which can reduce the interference of non-causal features on the classification labels then propose a novel Causality-inspired Domain Expansion Network (CDEN) for SDG. Specifically, CDEN contains a domain augmentation subnetwork and a representation learning subnetwork. The former uses a special style generator module AdaIN∗ to generate out-of-domain data with the same semantic information but different styles from the source domain data to help the model capture true associations between discriminative features and labels. The latter introduces a novel class-level contrastive learning, instance-level contrastive learning, and invariant risk minimization to help the model further eliminate spurious associations. CDEN progressively generates diverse style data and learns domain invariant representations containing few spurious associations by optimizing the two subnetworks in an adversarial learning manner. We conduct extensive experiments on three benchmark datasets to evaluate the superiority of CDEN, comparing it to fifteen state-of-the-art methods.

Systemic review of the Species Sensitivity Distribution Curve (CDSE) in environmental matrixes: a scientific contribution to ecotoxicological analysis

In ecotoxicity, differences in species sensitivity can be expressed through a curve based on the distribution of pre-existing contaminant toxicity data, enabling better comparative visualization and the deduction of taxonomic groups more and less sensitive to toxic agentes. The objective of this work was to carry out, in a structured way, an investigation in the literature about the application of the species sensitivity distribution (SSD) curve in ecotoxicological analyzes of environmental matrices. To assist in this process of knowledge construction, an adaptation of the intervention instrument ProKnow-C was used, in order to select relevant articles and identify the characteristics of these publications that will scientifically contribute to the theme in question. The main results are the identification of quantitative data for the year of publication, countries and most relevant journals aligned with the proposed theme, the minimum number of data for the construction of the curve and its main applications, as well as identifying gaps for new scientific research, thus contributing to the more precise application of the SDS curve in ecotoxicological analyses.

Class incremental learning with KL constraint and multi-strategy exemplar selection for classification based on MMFA model

Class incremental learning (CIL) can learn new classes continuously by updating the model rather than retraining a model from scratch with all seen classes like traditional offline learning, therefore, CIL is more suitable for classification in dynamic environments, where new classes are captured progressively. However, the key knowledge of old classes will be lost due to the update mode of CIL, leading to the catastrophic forgetting (CF) problem. In this paper, a novel CIL method with Kullback-Leibler constraint and multi-strategy exemplar selection (CIL-KLMES) is proposed for classification based on max-margin factor analysis (MMFA) model. To handle the CF problem, CIL-KLMES imposes a Kullback-Leibler (KL) divergence term on the important parameters when updating the model to restrict the parameters' distribution to be similar, thus preventing the updated model from deviating too much from the previous model and preserving the knowledge of old classes. Moreover, CIL-KLMES selects a few representative exemplars from old classes based on the robust description of data distribution and classification decision boundary. By replaying representative exemplars to update the model together with new class data, the key knowledge of old classes can be further preserved. Therefore, the CF problem can be alleviated sufficiently. Experimental results demonstrate the effectiveness of CIL-KLMES.