Skewness Research Articles

Fusion of multi-scale attention for aerial images small-target detection model based on PARE-YOLO

In view of the complex environments and varying object scales in drone-captured imagery, a novel PARE-YOLO algorithm based on YOLOv8 for small object detection is proposed. This model enhances feature extraction and fusion across multiple scales through a restructured neck network. Additionally, it incorporates a lightweight detection head that is optimized for small objects, thereby significantly improving detection performance in cluttered and intricate backgrounds. To further enhance the extraction of small object features, the conventional C2f is replaced with a novel architecture. Moreover, the EMA-GIoU loss function is proposed to mitigate class imbalance and enhance robustness, particularly in scenarios characterized by skewed class distributions. Evaluation on the VisDrone2019 dataset indicates that PARE-YOLO achieves a 5.9% improvement in mean Average Precision (mAP) at a threshold of 0.5, compared to the original YOLOv8 model. In addition, the PARE-YOLO model exhibits significant robustness, achieving a mean Average Precision (mAP) at a threshold of 0.5 values on the HIT-UAV dataset that are 0.8%, 0.5%, and 1.2% higher than those of YOLOv8, YOLOv10, and RT-DETR. These results underscore the effectiveness of PARE-YOLO in addressing the challenges inherent in aerial scenarios. The code will be available online (https://github.com/Sunnyxiao69/PARE-YOLO).

Warped multifidelity Gaussian processes for data fusion of skewed environmental data

Warped multifidelity Gaussian processes for data fusion of skewed environmental data

Genetic parameters and parental and early-life effects of boar semen traits

BackgroundThe objectives of this study were to estimate genetic parameters and studying the influence of early-life and parental factors on the semen traits of boars. The dataset included measurements on 449,966 ejaculates evaluated using a Computer-Assisted Sperm Analysis (CASA) system from 5692 artificial insemination (AI) boars. In total, we considered 16 semen traits measured on fresh semen and 6 sperm motility traits measured on semen after storage. Early-life effects included the dam’s parity, ages of the dam and sire, gestation length, litter size, litter sex ratio, number of piglets born alive, number of litter mates at weaning, rearing length, and weight gain. A repeatability model accounting for effects at collection was used to (1) estimate heritabilities and repeatabilities for semen traits and genetic and phenotypic correlations between traits, (2) test the significance of early-life effects, (3) quantify the contribution of exclusive dam and sire inheritances to the phenotypic variation, i.e., mitochondrial DNA and the Y chromosome, identified using a pedigree-based approach, and (4) quantify the contribution of maternal and paternal environment effects to the phenotypic variation of semen traits.ResultsWe reported heritabilities between 0.11 and 0.27 and repeatabilities between 0.20 and 0.65 for semen traits. Semen quality traits showed a skewed distribution, and their transformation significantly reduced their repeatability estimates. Motility traits measured after storage were genetically different from motility traits measured on fresh semen. Early-life had suggestive effects on a limited number of semen traits. Mitochondrial DNA and the Y chromosome did not explain a discerning proportion of the phenotypic variance and the effect of the paternal environment was also negligible. We estimated a significant maternal environment effect predominantly on sperm motility traits, explaining between 2.3 and 4.6% of the phenotypic variance. Including maternal environmental effects in the model reduced heritability estimates for sperm motility traits and total morphological abnormalities.ConclusionsOur findings indicate that trait transformation has a large effect on repeatability estimates of semen traits. Sperm motility traits measured on fresh semen are genetically different from sperm motility traits measured after storage. Early-life conditions can have an effect on later semen quantity and quality traits. Mitochondrial DNA and Y chromosome inheritances showed no effect on semen traits. Finally, we emphasize the importance of considering maternal effects when analysing semen traits, which results in lower heritability estimates.

Insertion/deletion polymorphism of the angiotensin-converting enzyme gene in lupus nephritis at Cho Ray hospital, Vietnam.

Introduction: The pathogenesis for poor kidney outcomes in lupus nephritis (LN) remains uncertain. There is limited evidence on the association between angiotensin-converting enzyme (ACE) I/D polymorphism and LN with kidney failure. Objectives: To determine the distribution of ACE I/D alleles and genotypes and the correlation between ACE I/D polymorphism and kidney outcomes in LN participants. Subjects and Methods: A cross sectional study was conducted at Cho Ray Hospital (12/2021-07/2023). The LN participants were stratified into 3 groups based on the kidney function and kidney replacement therapy (KRT) on admission. The ACE I/D polymorphism was identified by the polymerase chain reaction. Results: Among 208 LN participants, 71 were in group 1 (eGFR > 60), 55 in group 2 (eGFR ≤ 60 without KRT), 82 in group 3 (eGFR ≤ 60 with KRT). The skewed distribution among 3 groups were observed in D allele (20.4%, 35.5%, 37.2% in group 1, 2, 3, respectively, p = .003) and DD genotype (2.8%, 14.6%, 12.2% in group 1, 2, 3, respectively, p = .005). The ID/DD genotypes increased the susceptibility of kidney failure (eGFR ≤ 60 vs eGFR > 60: OR = 8.26 for DD genotype, OR = 2.31 for ID genotype) and KRT (KRT vs no KRT: OR = 2.01 for ID genotype). Conclusions: The D allele and ID/DD genotypes are linked with the susceptibility of kidney failure in LN participants.

Effectiveness of PEER Intervention on Older Adults’ Physical Activity Time Series Using Smoothing Spline ANOVA

Falls are a major cause of injury among older adults. The Physio-fEedback Exercise pRogram (PEER) combines physio-feedback, cognitive reframing, and guided exercises to reduce fall risk. However, its impact on physical activity (PA) over time is underexplored. Functional time-series analysis offers insight into behavior patterns and sustainability. This preliminary study assessed PEER’s effectiveness in improving PA levels immediately and over time. A total of 64 community-dwelling older adults were cluster-randomized into PEER (N=33) or control groups (N=31). Participants wore Fitbit trackers, generating time-series data on activity. The PEER group completed an 8-week program, while the control group received CDC fall prevention pamphlets. PA data were analyzed using smoothing spline analysis of variance (SSANOVA), chosen for its flexibility in modeling complex, non-linear relationships in time-series data and its ability to handle skewed distributions and repeated measures. Unlike traditional parametric models, SSANOVA decomposes temporal trends into interpretable components, capturing both smooth trends and abrupt changes, such as those occurring on group workout days. This capability ensures robust and nuanced analysis of intervention effects. Results showed PEER participants significantly increased evenly and had very active minutes and reduced sedentary behavior during the intervention. No significant effect was found for light active minutes. Specifically, during the intervention period, PEER participants engaged in an average of 6.7% fewer sedentary minutes per day, 13.8% additional fairly active minutes per day, and 2.8% additional very active minutes per day compared to the control group. While the reduction in sedentary minutes and increase in fairly active minutes were not statistically significant, the increase in very active minutes was significant. However, our functional time-series analysis revealed these improvements diminished over the 15-week follow-up, indicating challenges in maintaining PA. In conclusion, PEER boosts PA and reduces sedentary behavior short-term, but strategies are needed to sustain these benefits. In conclusion, PEER boosts PA and reduces sedentary behavior short-term, but strategies are needed to sustain these benefits. Public health policies should emphasize technology-driven fall risk assessments, community-based prevention programs, and initiatives that promote physical activity, home safety, and chronic condition management.

Pengaruh Kepemilikan Institusional, Kepemilikan Manajerial, Kinerja Keuangan, dan Women on Board Terhadap Pengungkapan Emisi Karbon

This study aims to examine the effect of institutional ownership, managerial ownership, financial performance, and women on board on carbon emission disclosure. The research utilize secondary data in the form of annual reports and sustainability reports of the companies. The population of this study consists of companies in the energy, infrastructure, and industrial sectors listed on the Indonesia Stock Exchange (IDX) for the 2021-2023 period. The data collection method employed in this study is purposive sampling with unbalanced data. A total of 147 companies data points met the criterias. The data analysis method used is quatitative, utilizing SPSS 30 for multiple linear regression analysis. The results of this study indicate that financial performance has a positive effect on carbon emission disclosure, while institutional ownership, managerial ownership, and women on board have no effect on carbon emission disclosure

Surficial Sediment Distribution in a Complex Marine Setting—The Example of Coastal and Open Sea Areas of Evia Island, Central Aegean, Greece

This study examines sediment distribution patterns in the Aegean Sea, focusing on the western coast of Evia Island and the southern Evoikos and Petalioi Gulfs. A total of 200 granulometric data points were analyzed to evaluate textural characteristics, including mean grain size, sorting, and skewness. The findings reveal significant variation in mean grain size: finer sediments (ϕ = 5–8) dominate the northern Aegean near Skyros, while coarser sands (ϕ = 2–4) are prevalent in the South Evoikos and Petalioi gulfs. The coarsest materials (ϕ = 0–2) are found around Chalkis, with sorting generally poor except in those two areas. Negative skewness values in the northern part of the study area indicate a predominance of finer particles, while southern regions exhibit slightly positive skewness, suggesting a greater proportion of coarser grains. Complex net transport patterns between Evia and Skyros are influenced by north-northeast trending water currents from the northern Aegean, with fluvial influx observed in the southern Evia coastal area. Around Skyros, the interplay of water currents and prevailing north-northwest winds dictates the orientation of net transport vectors. In the Evoikos gulfs, sediment movement aligns with a general northward flow, featuring coarse sediments in the Petalioi Gulf and muddy deposits in the narrower northern segment, where minimal transport indicates stagnant conditions. Coarse-grained materials in North Evoikos are primarily influenced by strong tidal activity.

Multi-perception Graph Convolutional Tree-embedded Network for Aero-engine Bearing Health Monitoring with Unbalanced Data

Multi-perception Graph Convolutional Tree-embedded Network for Aero-engine Bearing Health Monitoring with Unbalanced Data

Optimizing discrete dutch auctions with time considerations: a strategic approach for lognormal valuation distributions

This research deviates from usual studies in auction literature primarily focused on maximizing expected revenue. Instead, we concentrate on the strategic design of discrete Dutch auctions in the context of bidder emotional attachment, wherein valuations follow a lognormal distribution. Our objective is to attain an optimal balance between the auctioned object’s selling price and the auction duration, ultimately maximizing the auctioneer’s expected revenue per unit of time. Our proposed models exhibit significantly higher average revenues per unit of time than counterparts neglecting time considerations and emotional attachment of the bidders. This achievement results from strategically reducing auction durations, enabling more auctions within the allotted time. This intentional trade-off ensures the marginal revenue decrease in shorter auctions is surpassed by the substantial increase in overall revenues from heightened auction frequency. Numerical results emphasize the utility of our modified discrete Dutch auction design, particularly in scenarios with a large number of bidders. Furthermore, increasing skewness in valuation distributions correlates with higher revenue per unit of time. Complete knowledge of the number of participating bidders is crucial, leading to a noticeable elevation in the auctioneer’s expected revenue per unit of time. However, the predictability of auction outcomes may be challenging, underscoring the nuanced nature of auction dynamics.

Nonlinear relationship between blood urea nitrogen to albumin ratio and 3-month outcomes in patients with acute ischemic stroke: a second analysis based on a prospective cohort study

Patients with acute ischemic stroke (AIS) have limited evidence regarding the relationship between blood urea nitrogen and albumin. Aiming to investigate the relationship between the blood urea nitrogen to albumin ratio (BUN/ALB ratio) and poor outcomes in AIS patients at 3 months was the purpose of this study. AIS participants at a Korean hospital from January 2010 to December 2016 were included in a secondary analysis of a prospective cohort study. Logistic regression and restricted cubic splines were used to examine the relationship between BUN/ALB ratio and poor outcomes after 3 months. There is a skewed distribution of BUN/ALB ratios, ranging from 0.114 to 1.250. Model II of the binary logistic regression showed that the BUN/ALB ratio was not statistically significant in predicting poor outcomes for AIS patients after 3 months. However, there was a notable nonlinear relationship between them, with the inflection point of the BUN/ALB ratio identified as 0.326. The BUN/ALB ratio on the left side of the inflection point was associated with a 42% reduction in 3-month poor outcomes (odds ratio= 0.58, 95% confidence interval: 0.40 to 0.83). Conversely, the relationship was not statistically significant on the right side of the inflection point. The BUN/ALB ratio and poor outcomes in AIS patients show a nonlinear correlation. For AIS patients, a BUN/ALB ratio of approximately 0.326 is associated with the lowest risk of adverse outcomes at 3 months. Specifically, for nonsmoking AIS patients, a BUN/ALB ratio of approximately 0.295 is associated with the lowest risk of adverse outcomes at 3 months.

The growth and development of Nordic regional science research 1982–2022: bibliometric evidence from thirteen regional science journals

In a paper published in 2020, Philip McCann gives a qualitative overview of “the Nordic contribution to regional science”, which, however, only includes contributions by Swedish regional science researchers. The purpose of this paper is to complement McCann’s overview by highlighting Nordic contributions to regional science in the period 1983–2022 through a bibliometric analysis of publications in a set of thirteen regional science journals indexed in Web of Science, using several standard bibliometric tools. Our most interesting and surprising result is that the number of “Nordic” publications in the chosen set of journals has grown more than six times faster than the total number of publications in these journals between 1983–1992 and 2013–2022. This implies that the “Nordic” regional science researchers have increased their “market share” from 1.9 to 9.1 per cent. During the same period, the share of co-authored papers increased from 50.0 to 82.9 per cent and the share of international co-authorships increased from 0 to 50.0 per cent and went from being a mainly intra-European activity to a global activity. This process is also reflected in a certain international influence on the research topics during the four periods analysed, but to a considerable extent it seems that the Nordic regional scientists have pursued their own Nordic research themes. In terms of individual research productivity, there was the expected skewed distribution with a small number of researchers with a large research output. At the institutional level, there were notable changes in the ranking of institutions in terms of number of authorships, but one institution—Umeå University—ranked among the Nordic top-2 in all four periods.

Stigma of People Affected by Leprosy and its determinants: a cross-sectional study in Sri Lanka

BackgroundLeprosy, being a neglected tropical disease, remains a significant public health concern in several parts of the world, especially Sri Lanka, where it has long been associated with stigma. This study aims to assess the stigma faced by People Affected by Leprosy (PAL) undergoing treatment and to identify the factors associated with this stigma.MethodsA cross-sectional study was conducted among 109 Leprosy affected people attending the Central Leprosy Clinic and all dermatology clinics of the National Hospital of Sri Lanka (Colombo), selected using a consecutive sampling method. Data collection was done through an interviewer-administered questionnaire. Stigma was quantified using a validated and adapted version of the Stigma Assessment and Reduction of Impact (SARI) tool, which evaluates stigma pertaining to four domains: Experienced Stigma (ES), Disclosure Concerns (DC), Internalized Stigma (IS), and Anticipated Stigma (AS). Spearman’s correlation and Mann-Whitney U tests were used to analyze associations, with statistical significance set at 0.05.ResultsThe dataset showed a non-normal, right-skewed distribution. The mean total SARI score was 9.82 (SD = 10.23). Disclosure Concerns (rs=-2.66; p = 0.005) and Anticipated Stigma (rs=-3.6; p < 0.001) demonstrated weak - moderate negative correlations with the time since diagnosis. On the other hand, the correlation between time since diagnosis and Experienced Stigma (rs=0.22; p = 0.022) was positive and weak. Participants without disability had significantly lower scores in DC (p = 0.049), IS (p = 0.01), and AS (p = 0.01) compared to those with disability. Patients who did not develop Leprosy reactions displayed significantly lower scores across all SARI domains as well as the total SARI score (all p < 0.05) compared to those who did.ConclusionsThe findings suggest that over time, People Affected by Leprosy become less concerned about disclosing their diagnosis and anticipate less stigma but may experience greater overall stigma. Disability status and Leprosy reactions are key factors in determining the level of stigma faced by them.

Exploring Coding Sequence Length Distributions Across Taxonomic Kingdoms Based on Maximum Information Principle

Background: Genetic information about organisms' traits is stored and encoded in deoxyribonucleic acid (DNA) sequences. The fundamental inquiry into the storage mechanisms of this genetic information within genomes has long been of interest to geneticists and biophysicists. Objective: The objective of this study was to investigate the distribution of coding sequence (CDS) lengths in species genomes across different kingdoms. Methods: In this study, we used the maximum entropy principle and the gamma distribution model based on a comprehensive dataset including viruses, archaea, bacteria, and eukaryote species Results: Our study result revealed unique patterns in CDS length distributions among kingdoms and CDS lengths exhibit a right-skewed distribution, with varying preferences among kingdoms. Eukaryotes displayed bimodal distributions, with CDS sequences longer than those of prokaryotes. Fitting the gamma distribution model revealed differences in shape and scale parameters among kingdoms, with eukaryotes exhibiting larger scale parameters, indicating longer CDS sequences. Additionally, analysis of moments highlighted the complexity of eukaryotic genomes relative to prokaryotes. Conclusion: This study result deepens our understanding of genome evolution and provides valuable insights for biological research.Conclusion: This study result deepens our understanding of genome evolution and provides valuable insights for biological research.

Deep learning model based prediction of vehicle CO2 emissions with eXplainable AI integration for sustainable environment

The transportation industry contributes significantly to climate change through carbon dioxide (CO2) emissions, intensifying global warming and leading to more frequent and severe weather phenomena such as flooding, drought, heat waves, glacier melting, and rising sea levels. This study proposes a comprehensive approach for predicting CO2 emissions from vehicles using deep learning techniques enhanced by eXplainable Artificial Intelligence (XAI) methods. Utilizing a dataset from the Canadian government’s official open data portal, we explored the impact of various vehicle attributes on CO2 emissions. Our analysis reveals that not only do high-performance engines emit more pollutants, but fuel consumption under both city and highway conditions also contributes significantly to higher emissions. We identified skewed distributions in the number of vehicles produced by different manufacturers and trends in fuel consumption across fuel types. This study used deep learning techniques to construct a CO2 emission prediction model, specifically a light multilayer perceptron (MLP) architecture called CarbonMLP. The proposed model was optimized by hyperparameter tuning and achieved excellent performance metrics, such as a high R-squared value of 0.9938 and a low Mean Squared Error (MSE) of 0.0002. This study employs XAI approaches, particularly SHapley Additive exPlanations (SHAP), to improve the model interpretation ability and provide information about the importance of features. The findings of this study show that the proposed methodology accurately predicts CO2 emissions from vehicles. Additionally, the analysis suggests areas for further research, such as increasing the dataset, integrating additional pollutants, improving interpretability, and investigating real-world applications. Overall, this study contributes to the design of effective strategies for reducing vehicle CO2 emissions and promoting environmental sustainability.

Application of quantum-inspired tensor networks to optimize federated learning systems

Federated learning (FL) has gained significant traction across diverse industries, which allows multiple clients or institutions to enhance model performance and outcomes while preserving data privacy collaboratively. In recent years, tensor networks (TNs) have become important in machine learning because they allow the compact representation of high-dimensional tensors by decomposing them into lower-dimensional components with polynomial complexity. The application of TNs in FL is a natural extension because of its flexible framework for representing and optimizing models. Inspired by quantum computing principles, we have integrated a quantum-inspired tensor network into the FL framework. This framework focuses on a one-dimensional matrix product state (MPS) tensor network (TN) in a federated setting (FedTN), with data distributed across homogeneous and heterogeneous partitions among clients. Our experiments demonstrate that tensor network-based federated learning can be made practical, as FedTN is robust to the unbalanced and non-IID data distributions typically encountered in such settings. Our research assessed the effectiveness and feasibility of comparing quantum-inspired TN and conventional methods, evaluating their performance, and exploring the benefits of incorporating quantum principles in FL settings. Furthermore, we have investigated its performance when training for many local epochs (large E) between the averaging steps.

Advanced sleep disorder detection using multi-layered ensemble learning and advanced data balancing techniques

Sleep disorder detection has greatly improved with the integration of machine learning, offering enhanced accuracy and effectiveness. However, the labor-intensive nature of diagnosis still presents challenges. To address these, we propose a novel coordination model aimed at improving detection accuracy and reliability through a multi-model ensemble approach. The proposed method employs a multi-layered ensemble model, starting with the careful selection of N models to capture essential features. Techniques such as thresholding, predictive scoring, and the conversion of Softmax labels into multidimensional feature vectors improve interpretability. Ensemble methods like voting and stacking are used to ensure collaborative decision-making across models. Both the original dataset and one modified using the Synthetic Minority Oversampling Technique (SMOTE) were evaluated to address data imbalance issues. The ensemble model demonstrated superior performance, achieving 96.88% accuracy on the SMOTE-implemented dataset and 95.75% accuracy on the original dataset. Moreover, an eight-fold cross-validation yielded an impressive 99.5% accuracy, indicating the reliability of the model in handling unbalanced data and ensuring precise detection of sleep disorders. Compared to individual models, the proposed ensemble method significantly outperformed traditional models. The combination of models not only enhanced accuracy but also improved the system's ability to handle unbalanced data, a common limitation in traditional methods. This study marks a significant advancement in sleep disorder detection through the integration of innovative ensemble techniques. The proposed approach, combining multiple models and advanced interpretability methods, promises improved patient outcomes and greater diagnostic accuracy, paving the way for future applications in medical diagnostics.

The effects of skewness and kurtosis on production and hedging decisions: a Gram-Charlier expansion approach

In this study, we propose a Gram-Charlier expansion approach to investigate the impact of skewness and kurtosis on production and hedging decisions. Consistent with the existing literature, we find that skewness and kurtosis do not affect decisions regarding optimal production; however, they significantly influence optimal hedging decisions. We observe that positive skewness with platykurtic spot prices or negative skewness with leptokurtic spot prices often leads to over-hedging when the initial forward contract price exceeds its expected value. Conversely, under-hedging is expected when the initial forward contract price falls below its expected value. In other conditions, skewness can either promote or impede speculative future trading. Using the Gram-Charlier expansion of the spot price density function, we find that optimal future positions depend on forward prices, the hedgers’ risk preference, and the spot price distribution. Simulations validate our findings on the impact of skewness and kurtosis on future hedging. Finally, we analyze of a cotton storage and forward contracting dataset to illustrate the application of our methodology and support our theoretical results.

Sentiment Analysis of Movie and News: A Review on Unbalanced Data Using Deep Learning with Abstract Feature Extraction

This study assessed a sentiment analysis model using a BERT-based deep learning approach, combined with a feedforward neural network (FFNN), designed to handle unbalanced datasets. The model was evaluated on the Rotten Tomatoes movie review dataset and the 20 Newsgroups dataset. Our method achieved an outstanding performance, with perfect accuracy of 100% and minimal loss (0.001 training and 0.010 testing loss) on the Rotten Tomatoes dataset and similar results on the 20 Newsgroups dataset. Key metrics, such as the Jaccard index, Hamming loss, Cohen’s kappa, mean squared error, and Matthews correlation coefficient (MCC), demonstrated its effectiveness, particularly in managing class imbalance and capturing nuanced sentiment variations. Comparative analysis showed the proposed model to outperform existing methods, highlighting its robustness and potential application in sentiment classification tasks across various textual data domains.

Spread Regression, Skewness Regression, and Kurtosis Regression With an Application to the US Wage Structure

ABSTRACTQuantile regression provides a powerful tool for investigating the effects of covariates on key quantiles of a conditional distribution. However, we often lack a general picture of how covariates affect the overall shape of the conditional distribution. Using quantile regression estimation and quantile‐based measures of spread, skewness, and kurtosis, we propose spread regression, skewness regression, and kurtosis regression as empirical tools to quantify the effects of covariates on the spread, skewness, and kurtosis of the conditional distribution. This methodology is applied to US wage data during 1980–2019 with substantive findings, and a comparison is made with a moment‐based robust approach. In addition, we decompose changes in the spread into composition effects and structural effects to clarify rising inequality. We also provide the Stata commands spreadreg, skewreg, and kurtosisreg, which are available from the Statistical Software Components (SSC) archive, for easy implementation of spread, skewness, and kurtotis regressions.

Research on weight initialization of CNN student models based on knowledge distillation

Knowledge distillation is a process of weight compression where a complex teacher model trains a simplified student model, making the weights and their distribution crucial. This paper investigates the weight distribution in convolutional and fully connected layers of both teacher and student models. For convolutional layers, it was discovered that both teacher and student models exhibit a piecewise power-law distribution. A verification method based on the piecewise power law distribution of convolutional layers was proposed, and the correctness of this law was confirmed. Detailed analysis of the breakpoints and power exponents reveals that the teacher model has smaller breakpoints than the student model; for weights smaller than the breakpoint, the teacher model’s power exponent is lower than that of the student model, whereas, for weights larger than the breakpoint, the teacher model’s power exponent is higher. Based on these findings, a new weight initialization algorithm for convolutional layers was proposed. For fully connected layers, both models demonstrate a skewed distribution. A verification method based on the skewed distribution of fully connected layers was proposed, and the correctness of this law was confirmed. Analysis of the kurtosis and skewness indicates that the student model exhibits higher kurtosis and skewness than the teacher model. Based on these observations, a new weight initialization algorithm for fully connected layers was proposed. Experimental results show that both initialization methods improve the initial and final accuracy of the student model compared to the He initialization method.