Digital China Health (China)

Compute eXpress Link (CXL) memory disaggregation technology has emerged as a prominent solution within the realm of memory decoupling, characterized by its low latency, supporting for memory semantics and hardware-level cache coherency. It aims to break through the limitations of traditional CPU-centric architectures, driving data centers towards a more flexible and composable architecture to accommodate the rapid development of advanced heterogeneous computing applications, such as high-performance computing and large-scale Artificial Intelligence (AI) models. This survey reviews the development of CXL memory disaggregation technology across three generations of CXL protocols, associated with the evolution of representative functionalities and key concepts in CXL specifications. Meanwhile, the fundamental performance and heterogeneous interleaving capabilities of CXL memory on the latest Intel Xeon platform are disclosed and analyzed in this paper based on practical tests. Furthermore, the paper conducts a comprehensive assessment of the current software and hardware ecosystem surrounding CXL memory disaggregation, delineating its advantageous application scenarios and the technological trends in adjacent fields, as informed by collective academic and industrial experiences. Last, this paper concludes the opportunities and challenges for CXL memory disaggregation technology catering to emerging applications like AI from a commercialization standpoint, denoting its potential in the design of next-generation intelligent computing centers.

IntroductionCognitive impairments, especially in executive function, attention, and working memory, are prevalent in major depressive disorder (MDD). While studies has identified brain areas involved in these deficits, the connection between cognitive function and acoustic features of speech remains underexplored. This study examines this relationship in MDD.MethodsA longitudinal cohort study was conducted with 53 participants diagnosed with MDD (DSM-IV). Cognitive performance was assessed using the Trail Making Test (TMT), Digit Span (DS), Continuous Performance Test (CPT), and Symbol Coding (SC). Speech samples were analyzed for 6,373 vocal features across six domains: energy, pitch, prosody, spectral, voice quality, and others. Spearman’s correlation and principal component analysis (PCA) assessed the link between cognitive scores and speech features. Regression analyses explored the predictive value of speech features for cognitive performance.ResultsSignificant cognitive impairments in executive function and attention were observed. Strong correlations were found between spectral speech features, particularly high-frequency components, and sustained attention and executive function. PCA showed that depression severity independently predicted certain vocal feature components. Notably, speech features related to cognitive impairment remained consistent across different cognitive tasks, and treatment-induced changes were observed over 8 weeks.ConclusionsThis study demonstrates that speech features, particularly spectral components, reflect cognitive impairment in MDD, offering promising non-invasive biomarkers for monitoring cognitive function in depression. Larger studies with extended follow-up are needed to explore their potential in predicting cognitive outcomes in MDD.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12888-025-07243-y.

Under the large-scale integration of wind turbine and photovoltaic into the grid, the power system faces the challenge of insufficient flexibility for regulation. Coordinated planning of hydro-wind-solar-storage systems can effectively mitigate the output volatility of renewable energy sources. This paper proposes a distributionally robust planning method for hydro-wind-solar-storage systems based on the Wasserstein distance. First, taking into account the spatiotemporal correlations of factors such as wind speed and solar irradiance, an auxiliary classifier generative adversarial network (AC-GAN) is employed to generate a set of wind turbine and photovoltaic output scenarios. Then, a bilevel capacity planning model is constructed for the integrated system. The upper level aims to minimize investment costs by determining the optimal energy storage capacity, while the lower level focuses on minimizing operational costs through optimizing storage operation states and the output of various devices. Subsequently, an improved proximal policy optimization (PPO) algorithm, grounded in the Markov decision process framework, is used to solve the model. Finally, an actual case study based on a hydro-wind-solar system in Qinghai China is conducted to validate the effectiveness of the proposed method.

Focusing on the dual-modal analysis method of pilot operation sequences in typical emergencies scenarios, a real-time comparison mechanism based on the deviation amplitude of characteristic parameters is established by analyzing the mapping relationship between QAR engineering value parameters and SOP operation event database. This technical path verifies the feasibility of dynamic parameter matching in complex scenarios, laying the foundation for pilot operation evaluations that are subsequently extended to other typical event scenarios. This article will systematically explain the following core methods: pilot operation event sequence generation method, pilot operation monitoring and identification technology based on QAR data, and pilot operation deviation evaluation method. This will promote the in-depth development of pilot operation monitoring technology in multiple scenarios and full processes.

Background of study: Differentiated thyroid cancer (DTC) accounts for most thyroid malignancies and has favorable survival outcomes, yet up to 30% of patients experience recurrence, placing strain on follow-up systems in resource-limited settings. Conventional staging tools offer limited predictive precision. With increasing interest in machine learning (ML) for precision oncology, there is a need for interpretable, deployable models suitable for low-resource environments.Aims and scope of paper: To develop and validate an interpretable machine learning model for predicting thyroid cancer recurrence and assess its feasibility for deployment in constrained clinical settings, including African oncology contexts.Methods: A retrospective dataset of 383 DTC patients with at least 10-year follow-up was sourced from the UCI Machine Learning Repository. Thirteen demographic, clinical, and treatment-related predictors were included. Data preprocessing involved encoding, scaling, and class balancing using SMOTE. Logistic Regression, Random Forest, K-Nearest Neighbors, and Extreme Gradient Boosting (XGBoost) were trained with hyperparameter tuning via grid search and cross-validation. Performance was evaluated using accuracy, precision, recall, F1 score, and AUC-ROC.Result: XGBoost achieved the best performance with 97% accuracy, 95% recall, 94% precision, and an AUC-ROC of 0.93. The most influential predictors were age, smoking status, T and M staging, ATA risk category, and adenopathy. The final model was deployed as a browser-based decision support tool to enable real-time recurrence risk estimation.Conclusion: This study presents a high-performing and interpretable ML model for predicting DTC recurrence, demonstrating feasibility for use in low-resource oncology settings. External validation with African clinical datasets and integration into electronic health systems is recommended to enhance equity and clinical uptake.

In the field of CPR1000 core monitoring in nuclear power plants, the limitations of traditional analog signal processing technology are significant. In order to improve the monitoring level, this study constructs a monitoring model that integrates digital signal processing and machine learning. The study first analyzes the CPR1000 core monitoring system and related technical foundations, adopts data-driven and algorithm-level fusion strategies to design the model architecture, and optimizes it by improving the training algorithm and tuning hyperparameters. The experimental data comes from the CPR1000 core monitoring system and is used for model training and testing after cleaning and normalization preprocessing. The comparative experimental results show that the fusion model performs excellently, with a monitoring accuracy of 92 %, which is 10 % higher than that of a single machine learning model, and a mean square error of 0.03, which is 0.1 lower than that of the traditional analog signal processing system. In terms of identifying abnormal core conditions, the recall rate reaches <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0 \%}$</tex>, successfully identifying more than <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0 \%}$</tex> of abnormal conditions, and the false alarm rate is only 5 %. This shows that the fusion model can monitor the core status more accurately and efficiently, and provide strong technical support for the safe and stable operation of nuclear power plants. However, the resource consumption of this model under complex working conditions needs to be optimized, and subsequent research can focus on this.

To improve the efficiency of logistics distribution and balance the workload of drivers, this study introduces the standard deviation of workload time as an equilibrium measure and constructs a dual-objective mixed-integer programming model based on three-dimensional loading constraints to achieve the minimum total distance and workload balance. To solve the above model, a multi-stage hybrid algorithm is designed in this paper. Firstly, the clustering algorithm is used to effectively divide the distribution area, and then the non-dominated sorting genetic algorithm is used to solve the vehicle routing problem considering workload balance. At the same time, the constructive heuristic three-dimensional loading algorithm based on block loading is invoked to ensure the feasibility of cargo loading. Based on the historical data of a material flow distribution company, the feasibility of the constructed model and algorithm is verified. The research shows that the standard deviation of workload time can be reduced by 35.86%, and the average loading rate can be increased by 9.19% under the premise of increasing the distribution distance by only 1.6%. In addition, the proposed algorithm is compared with three other multi-objective optimization algorithms by constructing examples of different scales and using a series of standard example sets to verify its effectiveness and superiority.

This work considers the control energy problem for a class of large-scale networks built from smaller-scale factor networks through a new type of graph product, where the factor networks are simple and undirected, and described by Laplacian dynamics. The control energy of such network system is obtained, which can be quantized by the controllability Gramian-based metrics.

To assess the effectiveness of personalized embryo transfer (pET) in enhancing intrauterine pregnancy rates in women with polycystic ovary syndrome (PCOS) without recurrent implantation failure (RIF). This single-center, open-label, randomized controlled trial was conducted at a university-affiliated assisted reproductive medical center between October 2022 and May 2024. Eligible participants were randomly assigned to either the pET arm guided by the RNA-seq-based endometrial receptivity test (rsERT) or the regular frozen embryo transfer (FET) arm. Clinical outcomes from the first embryo transfer cycles following the tests were tracked and compared between the groups. A total of 121 patients were included and randomly assigned to either the FET arm (n = 60) or the pET arm (n = 61). Intention-to-treat analyses revealed no significant differences between the FET group and the pET group in terms of intrauterine pregnancy rate (61.2% vs. 60.0%, difference, - 1.2% [95% CI, - 20.2 to 17.9%], p = 0.901), embryo implantation rate (54.7% vs. 50.7%, difference, - 4.0% [95% CI, - 20.5 to 12.8%], p = 0.643), early miscarriage rate (3.3% vs. 6.7%, difference, 3.3% [95% CI, - 11.1 to 18.7%], p = 1.000), and ongoing pregnancy rate (59.2% vs. 56.0%, difference, - 3.2% [95% CI, - 22.3 to 16.2%], p = 0.749). Subgroup analyses of intrauterine pregnancy rates revealed no statistically significant differences between the groups, regardless of the analysis method used. Current evidence does not support the routine use of rsERT for personalized embryo transfer in PCOS patients without RIF. Further large-scale, well-designed clinical trials are needed. The trial was registered on the Chinese Clinical Trial Registry (registration no. ChiCTR2200064131, prospectively registered). Trial registration date: 2022-09-27.

Gas station sales exhibit significant volatility and nonlinear characteristics. Traditional time series forecasting models struggle to meet the practical demands of gas station sales prediction due to insufficient accuracy. This paper proposes a gas station sales forecasting model based on Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and an Attention mechanism. While incorporating objective influencing factors such as temperature, weather, fuel prices, day of the week, and holidays, the model also integrates gas station-specific features including promotions, competitor promotions, fuel-nonfuel interactions, and competitor interactions to aid prediction. Experimental results show that the model achieves an absolute percentage error within 8.4%, demonstrating its capability to deliver highly accurate results in large-scale forecasting tasks. This performance metric not only fulfills the requirements for efficient prediction but also underscores its substantial practical application value. Consequently, this model is well-suited for large-scale forecasting scenarios demanding high precision, providing a reliable guarantee for practical implementation.