This study, based on NIPT data from 1082 male fetuses, systematically analyzed the relationship between fetal Y chromosome concentration and factors such as gestational age and maternal BMI, and constructed a model to predict the optimal testing time. Results showed that Y chromosome concentration was positively correlated with gestational age (r=0.456, p<0.001) and negatively correlated with BMI (r=-0.234, p=0.002), suggesting that concentration increases with gestational age, while a higher BMI delays the achievement of effective concentrations. To address multicollinearity and outliers, a robust Bayesian ridge regression model was employed, and further comparisons were made with multiple linear regression, nonlinear models, and random forest methods. The random forest model demonstrated the best prediction performance (R²=0.9924, MAE=0.115 weeks), enabling individualized NIPT timing prediction with subweekly accuracy. This proposed model provides a scientific basis for optimizing NIPT testing time, reducing duplicate sampling, and improving test accuracy in clinical practice.

In order to achieve dynamical pattern recognition of univariate time series from the perspective of structural stability, this paper proposes a method based on Extended State Observer (ESO) and deterministic learning for addressing the issue of recognizing the structural stability of the topological structure of dynamical patterns in univariate time series. The ESO is capable of reconstructing the system states and their unknown dynamics during the observation phase, while the deterministic learning effectively resolves the challenges associated with obtaining and modeling the derivative information of the dynamics. During the learning phase, the ESO is employed to recover state trajectories from sampled output signals. These trajectories utilize the regressed trajectories as inputs, satisfying the partially persistent excitation (PE) condition, thereby accurately approximating the unknown dynamics through the estimated regressed trajectories. Subsequently, we establish a novel recognition error system that approximates the first-order derivatives of the system dynamics using a finite difference method, thus avoiding the need for re-modeling the test patterns and enabling the identification of dynamic behaviors from the topological structural perspective. Finally, we theoretically demonstrate that the residuals of the recognition error system reflect the structural stability differences in system dynamics between the training and test patterns, and simulation further substantiate the effectiveness and accuracy of the proposed method.

This paper presents a novel deep learning error minimization system designed to enhance the efficiency, adaptability, and accuracy of real-time big data analysis in mobile applications. Traditional deep learning systems face limitations such as the need for large labeled datasets, high computational complexity, and challenges in generalizing to new tasks and dynamic data. The proposed system addresses these limitations through a hybrid neural network architecture that integrates convolutional neural networks (CNNs) and recurrent neural networks (RNNs), a dynamic error feedback mechanism using reinforcement learning, and a lightweight transfer learning module. The system also includes modules for data preprocessing, augmentation, and model evaluation. Case studies demonstrate the system's effectiveness in real-time data analysis for applications such as financial market analysis, showcasing its potential to significantly improve performance and user experience in mobile applications.

The proliferation of large-scale language models, such as ChatGPT, has underscored the urgent requirement to develop Language Models in Medicine (LLMs) to mitigate the burden on healthcare resources. This work introduces MulMed, a model that prioritizes multitasking capabilities in medical domains. MulMed aims to summarize complex medical texts, address patient inquiries, engage in medical question-answering dialogues, demonstrate cross-lingual proficiency, and offer comprehensive medical knowledge coverage. Its key contributions include a two-step fine-tuned modeling framework that enables the model to perform multi-task functions like medical text summarization and Q&A in both English and Chinese, demonstrating excellent generalization abilities on benchmark test sets. The model also exhibits human empathy in doctor-patient consultations, and its fine-tuning process and data are openly available to promote future research in cross-lingual medical models. Additionally, a medical ethics framework is proposed to aid in evaluating the feasibility of medical model applications.

The aim of this paper is to enhance the performance of the target detection model in the insulator defect detection task. The YOLO v8 detection model is first improved by designing the Backbone-DCNv3 module, which is significantly enhanced in feature extraction and target localization capabilities. Meanwhile, the introduction of MPDIoU loss function further improves the detection accuracy of the model. Tests on the same dataset show that compared with the original model and other models in the YOLO series this paper's model exhibits superior performance in key evaluation metrics such as precision rate, recall rate and average precision. They for precision was 91.45%, recall was 89.47%, and mAP was 91.89%, respectively. Compared with the original model, the improvement was 2.13%, 2.02%, and 2.62%, respectively. Therefore, the research in this paper not only provides new ideas and methods in the field of target detection, but also provides strong technical support for the practical engineering application of insulator defect detection.

The swift advancement of generative artificial intelligence (GenAI), illustrated by tools such as ChatGPT, has garnered substantial interest regarding its potential applications across diverse fields. This research investigates the influence of GenAI on self-efficacy, employing transactional stress theory as a framework. Conceptualizing GenAI as a potential stressor, the research examines how challenge appraisals mediate its effects on self-efficacy. The empirical data show that perceiving GenAI as a learning and problem-solving tool boosts confidence in their academic capabilities. Conversely, negative attitudes toward GenAI can reduce its positive effects. These findings extend the application of transactional stress theory to modern technologies, offering insights for policymakers on promoting positive engagement with GenAI.

Temperature data inside the granary is a crucial indicator for safe storage and quality preservation of grain. Currently, temperature measurement relies on temperature cables installed inside the grain piles. However, the sensors on the cables can only acquire temperature data at specific measurement points and cannot provide a complete overview of the entire grain pile. Most current temperature field generation methods based on conventional spatial interpolation techniques often suffer from significant errors. To achieve a more intuitive and accurate monitoring of temperature distribution in granary environments, this study proposes a high-resolution reconstruction model for grain temperature fields based on the 3DSRCNN_Resnet. The model firstly improves the SRCNN in 3D to obtain the 3DSRCNN model, so as to apply the SRCNN super-resolution reconstruction algorithm, which deals with 2D images, to the task of reconstructing the temperature field of a three-dimensional granary. Secondly, a nonlinear mapping module based on 3D residual blocks and a residual reconstruction module based on 3D deconvolution are proposed to adjust the network structure of the 3DSRCNN model. The experimental results show that compared with the traditional spatial interpolation method, the grain temperature data reconstructed by this model achieves better performance under MSE, MAE, PSNR and SSIM metrics. This model provides a novel approach for comprehensive and accurate understanding of temperature distribution within grain piles.

Image style transfer is a popular research topic in the fields of computer graphics and multimedia, especially artistic stylization. The image style learning algorithm mainly studies the use of computer graphics and machine learning methods for automatic artistic rendering and intelligent processing of real sample data. The current mainstream methods mainly focus on learning static samples of artistic images. However, the information contained in static sample data is isolated and discontinuous; it is difficult to ensure the overall consistency of image style transfer. This article aims to recognize the real gesture movements of painters during the process of image style transfer, and propose a theoretical model and design method for image style transfer based on sequence task learning theory. Mainly completed the following research work: (1) Complete the design path and verification experiment of human-computer interaction gesture recognition in the air, use gesture recognition model, color space, color distribution model, color point probability, and maximum inter class variance method to complete the binary processing of images, and use Babbitt distance and learners to learn the intrinsic patterns of actions, ensuring that the gesture recognition design method in the air can be implemented; (2) Analyze and understand the real gesture process of painters, and design a dynamic decision-making model for image style transfer based on parameter exploration. Implement an image style transfer assistance system based on human-computer interaction gesture recognition.

This paper proposes an enhanced text matching model with augmented recurrent attention that utilizes interactive attention mechanisms. During vector encoding, the proposed model employs attention to interact between two input texts. Following the interaction, it leverages Bi-LSTM to re-encode the sequence at a more advanced level, enabling the model to comprehensively learn global information. Additionally, an attention mechanism is incorporated to emphasize the importance of high-weights words. Furthermore, a fusion layer is added to better integrate the two text segments into a single result, which facilitates subsequent text similarity computations. The model demonstrates a high accuracy in text similarity calculations.

The soil classification task has become a major challenge in the field of computer vision due to subtle differences between categories, scarce samples, and traditional classification methods that are time-consuming and labor-intensive. To solve this problem, this paper proposes a soil classification algorithm combining adaptive feature transformation (AFT) and dual metric module (CCE) to improve classification accuracy and generalization ability. The AFT module dynamically adjusts image features through affine transformation, enhances the model's adaptability to changes in feature distribution, and captures subtle differences. The CCE module combines the advantages of cosine similarity and Euclidean distance to improve the accuracy of feature similarity evaluation. The classification accuracy on the public data sets CUB-200-2011 and Stanford Dogs reached 87.36% and 81.50% respectively, and the classification accuracy on the soil data set Soil was 72.43%. Experimental results show that this method achieves significant performance improvement in few-shot soil classification tasks, verifying its effectiveness in complex classification problems. Future work will further explore the impact of local feature enhancement strategies on model performance to continuously optimize the classification effect.