Categorical Distribution Research Articles

Enhanced Student Performance Prediction Using Data Augmentation with Ensemble Generative Adversarial Network

Objective: To solve the issue of inadequate student data and increase the precision of performance prediction. Methods: This paper presents an innovative Ensemble Generative Adversarial Network (EGAN) model to augment student data and improve the accuracy of predicting student performance. The EGAN integrates two different GAN models: (i) Divergence GAN (DivGAN) and Success-aware GAN (SucGAN). The DivGAN reduces the difference between latent variables and observed data. The SucGAN uses Gumbel-Softmax Relaxation (GSR) to approximate the categorical distribution for creating high-quality data, which solves the imbalance ratio between raw and synthetic data in multiple classes (e.g., Pass, Fail, Distinction, and High Distinction). In contrast, only concentrating on data quality leads to mode collapse issue. So, a Tuna Swarm Optimization (TSO) is employed to stabilize the SucGAN training process and ensure a balance between data quality and diversity while training SucGAN. Thus, the synthetic student data is generated by EGAN and combined with the original dataset to form an augmented dataset. Moreover, this dataset is used by the Student Accomplishment prediction model using the Distinctive Deep Learning (SADDL) model to accurately predict student performance. Findings: The experimental findings illustrate that the EGAN-SADDL attains 94.71% accuracy and 0.0529 Root Mean Square Error (RMSE) in predicting student performance compared to the existing DL models. Novelty: This work expands the dataset by creating synthetic student data using EGAN. By increasing the amount of data available for training the SADDL model, this data augmentation strategy also boosts the model's accuracy and generalization abilities. Keywords: Student's performance prediction, SADDL, GAN, Divergence, Tuna swarm optimizer, Gumbel-Softmax relaxation

Novel R package to quantify and monetize burden of disease attributable to air pollution and noise

Abstract Background Ambient air pollution and noise substantially impact public health and the economy. Different methods to quantify these impacts are described in the literature. Objectives Within the European project BEST-COST (Burden of disease-based methods for estimating the socio-economic cost of environmental stressors), we aim to develop an open-source R package to quantify and monetise the disease burden attributable to air pollution and noise, including the impact of social inequalities. Methods The developed code will allow for different calculation pathways including: single baseline health data vs. age-specific estimates from life tables; relative and absolute risk; scenario comparison based on the population attributable fraction vs. population impact fraction; single-pollutant vs. correlated exposures; and outdoor vs. indoor air pollution. Moreover, the package can digest a variety of input data formats for the exposure-response function (e.g. relative risk with fixed shape vs. user-defined function) and exposure data (e.g. population-weighted mean vs. categorical distribution, spatial data). Eligible impact metrics will include morbidity incidence and prevalence, number of deaths, years of life lost, and years lived with disability. Finally, the R package will enable quantifying social inequities (using a novel multiple deprivation index) and monetising the attributable disease burden (using a state-of-the-art approach). Results During development, the R package will be tested in BEST-COST case studies in 5 European countries (BE, EE, FR, NO, PT), compared with existing tools, and discussed with key users in a workshop. The final R package will be available on GitHub in 2026. Conclusions This tool will allow the quantification and monetisation of the health impact of air pollution and noise. The R package will be a flexible programming resource that can be used as a standalone or combined with existing tools to build a more solid basis for evidence-based policy making. Key messages • This open-source tool will allow the quantification and monetisation of the health impact of air pollution and noise, including the impact of social inequalities. • The R package will be a flexible programming resource that can be used as a standalone or combined with existing tools to build a more solid basis for evidence-based policy making.

A Fast Nonparametric Sampling (NPS) Method for Time-to-Event in Individual-Level Simulation Models.

Individual-level simulation models often require sampling times to events, however efficient parametric distributions for many processes may often not exist. For example, time to death from life tables cannot be accurately sampled from existing parametric distributions. We propose an efficient nonparametric method to sample times to events that does not require any parametric assumption on the hazards. We developed a nonparametric sampling (NPS) approach that simultaneously draws multiple time-to-event samples from a categorical distribution. This approach can be applied to univariate and multivariate processes. We discretize the entire period into equal-length time intervals and then derived the interval-specific probabilities. The times to events can then be used directly in individual-level simulation models. We compared the accuracy of our approach in sampling time-to-events from common parametric distributions, including exponential, gamma, and Gompertz. In addition, we evaluated the method's performance in sampling age to death from US life tables and sampling times to events from parametric baseline hazards with time-dependent covariates. The NPS method estimated similar expected times to events from 1 million draws for the three parametric distributions, 100,000 draws for the homogenous cohort, 200,000 draws from the heterogeneous cohort, and 1 million draws for the parametric distributions with time-varying covariates, all in less than a second. Our method produces accurate and computationally efficient samples for time-to-events from hazards without requiring parametric assumptions.

A novel tree-based method for interpretable reinforcement learning

Deep reinforcement learning (DRL) has garnered remarkable success across various domains, propelled by advancements in deep learning (DL) technologies. However, the opacity of DL presents significant challenges, limiting the application of DRL in critical systems. In response, decision tree (DT)-based methods, known for their transparent decision-making mechanisms, have shown promise in making interpretable policies for decision-making problems. Existing methods often employ differential DTs to model RL policies and discretize them to conventional DTs for higher interpretability. Yet, this method leads to discrepancies between the trained differential DTs and the discretized DTs. To address this issue, we introduce Generative Consistent Trees (GCTs), a novel solution that circumvents the information loss typically associated with the argmax operation in prior research. By implementing a reparameterization technique to approximate the categorical distribution, GCTs ensure the consistencies between trained GCTs and discretized counterparts. Moreover, we have developed an imitation-learning-based framework for interpretable reinforcement learning. This framework is designed to train GCTs by efficiently mimicking expert policies. Our extensive experiments across multiple environments have validated the effectiveness of this approach, highlighting the potential of GCTs in enhancing the interpretability and applicability of DRL.

Dimensional Distribution of Anxiety and Depression in College Students in a Rural Setting: Relationship with Stress, Well-being, and Quality of Life

Background: Information about the anxiety and depression of students in rural areas of India is scant. We evaluated stress, coping strategies, quality of life (QOL), and well-being of students along with the dimensional and categorical distribution of anxiety and depression in a cross-sectional study in a rural college. Materials and Methods: It was an online questionnaire-based survey using the Patient Health Questionnaire-9 for depression, general anxiety disorder-7 for anxiety, and the WHO-5 Well-Being scale for well-being. Stress in the previous month, nature of stress, coping strategies, and QOL on a scale of 1 (worst) to 10 (best) were also assessed. Results: The dimensional distribution of anxiety and depression suggested the pervasiveness of these problems among students. Categorically, a considerable proportion of students had clinical levels of anxiety (32.2%), depression (40.3%); and 24.8% of students had both of them. Stress in the past month was reported by 34.2% of students, which was linked significantly with the severity of anxiety and depression. Common stressor were financial problems, family-, career-, and job-related issues. Stress, anxiety, and depression were negatively correlated with well-being and QOL. Many coping strategies were reported; more than half of the students found helping others, creative hobbies, hoping for the best, music, talking to others, yoga and exercises, prayer, and religious activities helpful. Conclusions: The results highlight the pervasive presence of anxiety and depression in college students, which are linked to stress, and have an impact on their well-being and QOL. These require specific attention for management.

Bayesian improved cross entropy method with categorical mixture models for network reliability assessment

We employ the Bayesian improved cross entropy (BiCE) method for rare event estimation in static networks and choose the categorical mixture (CM) as the parametric family to capture the dependence among network components. The proposed method is termed BiCE-CM. At each iteration of BiCE-CM, the mixture parameters are updated through the weighted maximum a posteriori (MAP) estimate, which mitigates the overfitting issue of the standard improved cross entropy (iCE) method through a novel balanced prior, and we propose a generalized version of the expectation–maximization (EM) algorithm to approximate this weighted MAP estimate. The resulting importance sampling distribution is proved to be unbiased. For choosing a proper number of components K in the mixture, we compute the Bayesian information criterion (BIC) of each candidate K as a by-product of the generalized EM algorithm. The performance of the proposed method is investigated through a simple illustration, a benchmark study, and a practical application. In all these numerical examples, the BiCE-CM method results in an efficient and accurate estimator that significantly outperforms the standard iCE method and the BiCE method with the independent categorical distribution.

Deep Subdomain Learning Adaptation Network: A Sensor Fault-Tolerant Soft Sensor for Industrial Processes.

Sensor faults are non-negligible issues for soft sensor modeling. However, existing deep learning-based soft sensors are fragile and sensitive when considering sensor faults. To improve the robustness against sensor faults, this article proposes a deep subdomain learning adaptation network (DSLAN) to develop a sensor fault-tolerant soft sensor, which is capable of handling both sensor degradation and sensor failure simultaneously. Primarily, domain adaptation works for process data with sensor degradation in industrial processes. Being founded on the basic structure of deep domain adaptation, a novel subdomain learner is added to automatically learn the subdomain division, enabling DSLAN adaptable to multimode industrial processes. Notably, the subdomain structure of each sample follows a categorical distribution parameterized by output of the subdomain learner. Based on the designed subdomain learner, a new probabilistic local maximum mean discrepancy (PLMMD) is presented to measure the difference in distribution between source and target features. In addition, a generator for failure data imputation is integrated in the framework, making DSLAN handle sensor failure simultaneously. Finally, the Tennessee Eastman (TE) benchmark process and two real industrial processes are used to verify the effectiveness of the proposed method. With the fault tolerance ability, soft sensing technology will take a step toward practical applications.

Traumatic Dental Injuries and their Association with Demographics and Other Predisposing Risk Factors in School Children Aged between 8-13 Years: A Descriptive, Cross-Sectional Study.

Traumatic dental injuries are most commonly seen in school age children. This not only affects physical, but, also, the psychological and social well-being of a child. The aim of the present study was to assess the prevalence of traumatic dental injuries in school children aged between 8 and 13 years and their association with the demographics and other predisposing risk factors. The present study was conducted as a cross-sectional study design on 1500 school children aged between 8 and 13 years. Apart from clinical examination, all the participants were asked to complete a questionnaire related to the injuries sustained. Statistical analysis was conducted using Statistical Package for Social Sciences (SPSS) version 21.0 (IBM, Chicago). Pearson's Chi-square (Χ²) test was used for categorical data analysis and distribution to determine the association between categorical variables, while P < 0.05 was considered statistically significant. The prevalence of traumatic dental injuries in the present study was calculated at 9.7%. Also, the occurrence of such injuries was found to be higher in the age group of 8-11 years with 10 years being the peak age of incidence for sustaining such injuries. Again, crown fracture involving enamel was found to be the most common type of injury sustained, while proclination observed in relation to upper anterior teeth, and incompetent lips were found to be the independent, significant risk predictors associated with the frequency and severity of such injuries. The results obtained in the present study suggested a high prevalence of traumatic dental injuries in school children aged between 8 and 13 years highlighting the need for conducting educational programs, for increasing knowledge and awareness of parents, school children and school faculty in this regard to save the affected child patients from the more dreaded sequelae including complete loss of tooth structure, and social and emotional trauma.

Quality evaluation of internet resources related to breast surgery in major search engines in Korea

Background With the increasing number of internet users, search engines have become a widely used source of health-related information. However, evaluating the quality of medical information obtained through search engines can be challenging. This study aimed to evaluate the quality of search results related to breast implants obtained from major domestic websites in Korea using systematic evaluation standards.Methods Two main primary search terms (“gaseum” [breast] implant and “yubang” [breast] implant), along with 15 adjunct search terms, were combined and searched in Google, Naver, and Daum. The top 20 websites were evaluated and classified according to their type and provider. They were scored using the Korean Medical Association’s Internet Health Information Certification Standards. The proportion of significant websites, their categorical distribution, and the quality of information scores were then compared.Results Google yielded the highest number of appropriate results, with statistical significance. Blogs (36.4%) and news (34.8%) were the most common types of search results, while healthcare provider groups (49.5%) constituted the most common provider subgroup. Only 1.9% of the search results were from public organizations. Google had a significantly higher average quality score (14.04) than Naver (13.22), and Daum (12.45) (P&lt;0.05).Conclusions Although almost half of the search results were provided by medical personnel, their average quality score (13.16) was below the overall average (13.36) and far below the 21 points of the journal/abstract category. The findings highlight the need for healthcare providers to provide high-quality medical information, and for users to develop high-level digital health literacy.

Modeling the complexity of basketball games using marked mutually exciting point processes

The basketball game is rarely predictable due to the complexity of its scoring process. To deal with the complexity and forecast the game results, we develop a marked mutually exciting point process model for the scoring process. In the model, the dynamic of the scoring intensity is characterized by a mutually exciting point process. A time varying background rate is used in the intensity function to account for the non-homogeneity. The dependence between scoring events is incorporated into the model with two team-specific exponential exciting kernels. The distribution of the points obtained in each scoring event is modeled as a categorical distribution with parameters from a Dirichlet prior. An empirical Bayesian method and an expectation-maximization algorithm are developed to estimate the model parameters. We design a simulation algorithm to do pre-game and in-game predictions. An empirical study is conducted with National Basketball Association games in four seasons. It is showed that the proposed model outperforms the benchmark model both in the win-loss prediction and score prediction. Moreover, our model can obtain positive returns when betting in the point spread market. Besides forecasting, an index called added winning probability is designed based on the model to evaluate the players’ performance.

How to Trade Off the Quantity and Capacity of Teacher Ensemble: Learning Categorical Distribution to Stochastically Employ a Teacher for Distillation

We observe two phenomenons with respect to quantity and capacity: 1) more teacher is not always better for multi-teacher knowledge distillation, and 2) stronger teacher is not always better for single-teacher knowledge distillation. To trade off the quantity and capacity of teacher ensemble, in this paper, we propose a new distillation paradigm named Dynamic Knowledge Distillation (DynaKD) that learn an adaptive categorical distribution to stochastically employ a teacher from a teacher ensemble in each step, to transfer knowledge from teacher ensemble into student. DynaKD has three advantages: 1) it can preserve diversity of each teacher via one-to-one distillation manner instead of several-for-one, 2) it can make the best of powerful teacher via those multi-level assistant teachers in ensemble, and 3) it can also dynamically determine the importance of each teacher for various tasks. To verify the effectiveness of the proposed approach, we conduct extensive experiments for BERT compression on GLUE benchmark. Experimental results show that the proposed approach achieves state-of-the-art score compared to previous compression approaches on five out of seven downstream tasks, including pushing MRPC F1 and accuracy to 92.2 (1.4 point absolute improvement), RTE accuracy to 76.2 (2.8 point absolute improvement). Moreover, we conduct also extensive experiments for image classification on CIFAR-100. Similarly, DynaKD achieves also state-of-the-art performance.

Towards Evidential and Class Separable Open Set Object Detection

Detecting in open-world scenarios poses a formidable challenge for models intended for real-world deployment. The advanced closed set object detectors achieve impressive performance under the closed set setting, but often produce overconfident misprediction on unknown objects due to the lack of supervision. In this paper, we propose a novel Evidential Object Detector (EOD) to formulate the Open Set Object Detection (OSOD) problem from the perspective of Evidential Deep Learning (EDL) theory, which quantifies classification uncertainty by placing the Dirichlet Prior over the categorical distribution parameters. The task-specific customized evidential framework, equipped with meticulously designed model architecture and loss function, effectively bridges the gap between EDL theory and detection tasks. Moreover, we utilize contrastive learning as an implicit means of evidential regularization and to encourage the class separation in the latent space. Alongside, we innovatively model the background uncertainty to further improve the unknown discovery ability. Extensive experiments on benchmark datasets demonstrate the outperformance of the proposed method over existing ones.

Digital Media Use and Health Literacy Levels of Women in Turkey

ABSTRACT Obtaining information about women’s use of digital media, their behavior in acquiring health information in the digital environment and their level of health literacy will fill an important gap in the literature. This research was conducted to examine the digital media use and health literacy levels of women in Turkey. The population of the research was the female population between the ages of 20–60 in Turkey in 2020, and the sample consisted of 404 women in this age range. The data of this analytical-cross-sectional study were collected online with the Personal Information Form and the Turkish Health Literacy Scale-32. It was determined that 83.9% of the women participating in the study used digital media and 82.9% of them obtained health information in the digital environment. In the categorical score distribution of the scale, it was determined that the health literacy level of 48.8% of women was insufficient, 26.2% problematic/limited, 19.8% sufficient and 5.2% excellent. A significant relationship was determined between women’s health literacy level and age, marital status, education level, employment status, perception of income level, place of residence, use of digital media and duration of use (p < .05). We concluded that majority of women in Turkey use digital media, obtain health information in the digital environment and have low health literacy levels.

A counting-time optimization method for artificial neural network (ANN) based gamma-ray spectroscopy

With advancements in machine learning technologies, artificial neural networks (ANNs) are being widely used to improve the performance of gamma-ray spectroscopy based on NaI(Tl) scintillation detectors. Typically, the performance of ANNs is evaluated using test datasets composed of actual spectra. However, the generation of such test datasets encompassing a wide range of actual spectra representing various scenarios often proves inefficient and time-consuming. Thus, instead of measuring actual spectra, we generated virtual spectra with diverse spectral features by sampling from categorical distribution functions derived from the base spectra of six radioactive isotopes: 54Mn, 57Co, 60Co, 134Cs, 137Cs, and 241Am. For practical applications, we determined the optimum counting time (OCT) as the point at which the change in the Kullback–Leibler divergence (ΔKLDV) values between the synthetic spectra used for training the ANN and the virtual spectra approaches zero. The accuracies of the actual spectra were significantly improved when measured up to their respective OCTs. The outcomes demonstrated that the proposed method can effectively determine the OCTs for gamma-ray spectroscopy based on ANNs without the need to measure actual spectra.

The neural substrates of transdiagnostic cognitive-linguistic heterogeneity in primary progressive aphasia

BackgroundClinical variants of primary progressive aphasia (PPA) are diagnosed based on characteristic patterns of language deficits, supported by corresponding neural changes on brain imaging. However, there is (i) considerable phenotypic variability within and between each diagnostic category with partially overlapping profiles of language performance between variants and (ii) accompanying non-linguistic cognitive impairments that may be independent of aphasia magnitude and disease severity. The neurobiological basis of this cognitive-linguistic heterogeneity remains unclear. Understanding the relationship between these variables would improve PPA clinical/research characterisation and strengthen clinical trial and symptomatic treatment design. We address these knowledge gaps using a data-driven transdiagnostic approach to chart cognitive-linguistic differences and their associations with grey/white matter degeneration across multiple PPA variants.MethodsForty-seven patients (13 semantic, 15 non-fluent, and 19 logopenic variant PPA) underwent assessment of general cognition, errors on language performance, and structural and diffusion magnetic resonance imaging to index whole-brain grey and white matter changes. Behavioural data were entered into varimax-rotated principal component analyses to derive orthogonal dimensions explaining the majority of cognitive variance. To uncover neural correlates of cognitive heterogeneity, derived components were used as covariates in neuroimaging analyses of grey matter (voxel-based morphometry) and white matter (network-based statistics of structural connectomes).ResultsFour behavioural components emerged: general cognition, semantic memory, working memory, and motor speech/phonology. Performance patterns on the latter three principal components were in keeping with each variant’s characteristic profile, but with a spectrum rather than categorical distribution across the cohort. General cognitive changes were most marked in logopenic variant PPA. Regardless of clinical diagnosis, general cognitive impairment was associated with inferior/posterior parietal grey/white matter involvement, semantic memory deficits with bilateral anterior temporal grey/white matter changes, working memory impairment with temporoparietal and frontostriatal grey/white matter involvement, and motor speech/phonology deficits with inferior/middle frontal grey matter alterations.ConclusionsCognitive-linguistic heterogeneity in PPA closely relates to individual-level variations on multiple behavioural dimensions and grey/white matter degeneration of regions within and beyond the language network. We further show that employment of transdiagnostic approaches may help to understand clinical symptom boundaries and reveal clinical and neural profiles that are shared across categorically defined variants of PPA.

Neural Categorical Priors for Physics-Based Character Control

Recent advances in learning reusable motion priors have demonstrated their effectiveness in generating naturalistic behaviors. In this paper, we propose a new learning framework in this paradigm for controlling physics-based characters with improved motion quality and diversity over existing methods. The proposed method uses reinforcement learning (RL) to initially track and imitate life-like movements from unstructured motion clips using the discrete information bottleneck, as adopted in the Vector Quantized Variational AutoEncoder (VQ-VAE). This structure compresses the most relevant information from the motion clips into a compact yet informative latent space, i.e., a discrete space over vector quantized codes. By sampling codes in the space from a trained categorical prior distribution, high-quality life-like behaviors can be generated, similar to the usage of VQ-VAE in computer vision. Although this prior distribution can be trained with the supervision of the encoder's output, it follows the original motion clip distribution in the dataset and could lead to imbalanced behaviors in our setting. To address the issue, we further propose a technique named prior shifting to adjust the prior distribution using curiosity-driven RL. The outcome distribution is demonstrated to offer sufficient behavioral diversity and significantly facilitates upper-level policy learning for downstream tasks. We conduct comprehensive experiments using humanoid characters on two challenging downstream tasks, sword-shield striking and two-player boxing game. Our results demonstrate that the proposed framework is capable of controlling the character to perform considerably high-quality movements in terms of behavioral strategies, diversity, and realism. Videos, codes, and data are available at https://tencent-roboticsx.github.io/NCP/.

Towards Improved Remedial Outcomes in Categorical Aquifers with an Iterative Ensemble Smoother.

Categorical parameter distributions consisting of geologic facies with distinct properties, for example, high-permeability channels embedded in a low-permeability matrix, are common at contaminated sites. At these sites, low-permeability facies store solute mass, acting as secondary sources to higher-permeability facies, sustaining concentrations for decades while increasing risk and cleanup costs. Parameter estimation is difficult in such systems because the discontinuities in the parameter space hinder the inverse problem. This paper presents a novel approach based on Traveling Pilot Points (TRIPS) and an iterative ensemble smoother (IES) to solve the categorical inverse problem. Groundwater flow and solute transport in a hypothetical aquifer with a categorical parameter distribution are simulated using MODFLOW 6. Heads and concentrations are recorded at multiple monitoring locations. IES is used to generate posterior ensembles assuming a TRIPS prior and an approximate multi-Gaussian prior. The ensembles are used to predict solute concentrations and mass into the future. The evaluation also includes an assessment of how the number of measurements and the choice of the geological prior determine the characteristics of the posterior ensemble and the resulting predictions. The results indicate that IES was able to efficiently sample the posterior distribution and showed that even with an approximate geological prior, a high degree of parameterization and history matching could lead to parameter ensembles that can be useful for making certain types of predictions (heads, concentrations). However, the approximate geological prior was insufficient for predicting mass. The analysis demonstrates how decision-makers can quantify uncertainty and make informed decisions with an ensemble-based approach.

Multi-view dreaming: multi-view world model with contrastive learning

In this paper, we propose Multi-View Dreaming, a novel reinforcement learning agent for integrated recognition and control from multi-view observations by extending Dreaming. Most current reinforcement learning method assumes a single-view observation space, and this imposes limitations on the observed data, such as lack of spatial information and occlusions. This makes obtaining ideal observational information from the environment difficult and is a bottleneck for real-world robotics applications. In this paper, we use contrastive learning to train a shared latent space between different viewpoints and show how the Products of Experts approach can be used to integrate and control the probability distributions of latent states for multiple viewpoints. We also propose Multi-View DreamingV2, a variant of Multi-View Dreaming that uses a categorical distribution to model the latent state instead of the Gaussian distribution. Experiments show that the proposed method outperforms simple extensions of existing methods in a realistic robot control task.

Generating evidential BEV maps in continuous driving space

Safety is critical for autonomous driving, and one aspect of improving safety is to accurately capture the uncertainties of the perception system, especially knowing the unknown. Different from only providing deterministic or probabilistic results, e.g., probabilistic object detection, that only provide partial information for the perception scenario, we propose a complete probabilistic model named GevBEV. It interprets the 2D driving space as a probabilistic Bird’s Eye View (BEV) map with point-based spatial Gaussian distributions, from which one can draw evidence as the parameters for the categorical Dirichlet distribution of any new sample point in the continuous driving space. The experimental results show that GevBEV not only provides more reliable uncertainty quantification but also outperforms the previous works on the benchmarks OPV2V and V2V4Real of BEV map interpretation for cooperative perception in simulated and real-world driving scenarios, respectively. A critical factor in cooperative perception is the data transmission size through the communication channels. GevBEV helps reduce communication overhead by selecting only the most important information to share from the learned uncertainty, reducing the average information communicated by 87% with only a slight performance drop. Our code is published at https://github.com/YuanYunshuang/GevBEV.

A prospective longitudinal cohort study of the effectiveness of 25% xylitol toothpaste on mutans streptococci in high caries-risk young children.

To compare the prevalence of mutans streptococci (SM) prevalence and categorical distribution in the saliva of high caries-risk children after their 24 months exposure to 25% xylitol toothpaste applied twice a day. To also examine the early caries susceptibility at baseline. An observational design with two times of collection. A total of 270 mother-infant dyads (mean age of infant 6.7 months) were initially recruited. At baseline, a dental exam looked for any early clinical signs of caries, and saliva collections were performed to assess SM prevalence. Two years later, other saliva collections were performed on a residual sample of 102 young children at the endpoint. Two hundred four saliva samples were analysed with Dentocult, a commercial kit estimating in four categories, the SM distribution, expressed as the number of colonies forming units per milliliter of saliva (CFU/ml). The dental exam used the ICDAS system to include initial carious lesions (d1-2). The dentist administrated a maternal socio-economic questionnaire as the study collector. A domestic strategy using a natural toothpaste containing 25% of xylitol as a vehicle, and tooth brushing as a habit, showed a promising antibacterial cariogenic effect. This approach could be a relevant alternative to in-home care to prevent ECC and early SM contamination.