Beijing Administration Institute

The reality gap between simulation and real-world dynamics critically hinders the deployment of robust humanoid locomotion policies, as policies trained in a single simulator often overfit to domain-specific dynamics. To address this challenge, we propose CROSSER (Inverse Dynamics-Guided Cross-Simulator Adaptation), a reinforcement learning framework that synergizes heterogeneous simulators by leveraging multi-source data distributions to train policies resilient to cross-simulator dynamic discrepancies. Unlike prior methods that naively train policies from a specific simulator, CROSSER employs an inverse dynamics model trained across diverse simulator trajectories. By quantifying cross-simulator action-space inconsistencies, it can effectively identify and prioritize states where domain-specific biases significantly degrade generalization. Through dynamically penalizing or filtering transitions according to these inconsistencies, CROSSER guides policies to learn unified strategies that harmonize divergent physical priors embedded in multi-simulator data, rather than overfitting to any single source. Our work establishes a novel paradigm for transforming cross-simulator discrepancies into actionable signals through inverse dynamics models, advancing the practical deployment of humanoid robots via collaborative multi-simulator training. Experiments on humanoid locomotion tasks demonstrate that CROSSER significantly enhances cross-simulator performance compared to single-simulator baselines and outperforms existing joint training methods in the locomotion task, while presenting more stable locomotion behaviors.

Cultural instruction plays a vital role in international Chinese language education; however, the systematic integration of cultural content into curricula and teaching materials remains underexplored at the empirical level. This study addresses this gap by investigating learners’ cognitive understanding of Chinese cultural vocabulary, aiming to support more targeted cultural instruction and curriculum development across diverse learner populations. Employing a corpus-based methodology, the study constructed a specialized corpus of 2329 culturally significant words aligned with international Chinese education standards. From this corpus, 93 representative words were selected across elementary, intermediate, and advanced levels to develop a Likert scale-based assessment tool. The tool’s psychometric properties were validated through empirical testing with 255 learners from various cultural backgrounds. Results indicate that learners’ Chinese language proficiency, duration of study, and cultural background significantly influence their mastery of cultural vocabulary. The study offers a novel and validated instrument for measuring cultural vocabulary knowledge, laying the groundwork for data-driven cultural syllabus design and instructional strategies. These findings contribute to enhancing intercultural competence among Chinese language learners and provide a scalable framework for advancing cultural instruction in international Chinese language education.

Against the backdrop of the UK’s severe educational stratification—where private schools educate only 7% of pupils but account for 39% of high elites, while public education suffers from resource shortages—this study examines the socioeconomic mobility effects of the UK Labor Party’s proposal to impose a 20% Value-Added Tax (VAT) on private school tuition fees. The research employs a combination of theoretical frameworks (including Bourdieu’s cultural capital theory, Rawls’ “Difference Principle,” and Mincer’s human capital theory) and empirical data (from sources such as the Social Mobility Commission, Ofsted, and ONS) to analyze how VAT affects private educational institutions, families across income groups, and long-term social and economic mobility. The research purposes are to clarify the transmission mechanism of VAT’s impact on educational resource allocation, identify its heterogeneous effects on different social classes, and provide evidence-based policy recommendations to mitigate educational inequality. The research results indicate that: 1) Imposing VAT significantly alters the cost structure of private schools—schools either pass on costs (leading to a 20% average tuition increase) or cut expenses (reducing extracurricular offerings and teacher training, which lowers the teaching quality index by 21%); 2) Family responses show obvious class differentiation: the top 10% income group maintains private education participation (strengthening intergenerational advantage, with elite closure probability rising by 37%), 53% of the middle 40% switch to public schools (disrupting cultural capital inheritance, with career promotion probability falling by 29%), and 82% of the bottom 50% enter public or informal education (worsening intergenerational educational poverty, with social bottom solidification rate rising by 41%); 3) In the long run, VAT exacerbates educational opportunity inequality (forming a “class fault in educational opportunity”) and career-income disparities (private school graduates are 42% more likely to enter high-income sectors like law and finance, with median lifetime income 39.6% higher than public school peers), ultimately consolidating class divisions and weakening the public education system. The study concludes that the current 20% VAT on private school tuition fees intensifies socioeconomic stratification. It recommends implementing progressive VAT exemptions (e.g., tiered taxation for low-fee private schools), increasing direct funding for public schools, and learning from Sweden’s education voucher system to balance tax efficiency, market function, and social fairness.

The pervasive use of plastic products has led to environmental contamination by compounds like 1,3-diphenylpropane (SD-1), a polystyrene dimer found in plastic food containers that poses potential health risks. SD-1 can induce coagulation disorder, however, the toxic mechanisms of SD-1 has not been elucidated yet. This study proposes a network toxicology analysis framework-InfomaxNet, which successfully addresses the challenge of lacking prior biological knowledge by analyzing complex biological networks using only network topology. Using the deep learning model MolTrans to predict SD-1 targets, InfomaxNet identified the critical proteins AKT2 and F9. Molecular dynamics simulations revealed that the binding of SD-1 to F9 (FIXa) induces conformational anomalies in its active site, disrupting protein function and increasing the risk of coagulation disorders. In vitro experiments confirmed that SD-1 interferes with coagulation pathways involving F9. Subsequently, acute toxicity experiments in Caenorhabditis elegans and RT-qPCR validated the impact of SD-1 on AKT2 and its downstream signaling pathways. This study introduces the InfomaxNet framework and applies it to network toxicology analysis, combining deep learning and molecular dynamics simulations to uncover the toxic mechanisms of SD-1 on the coagulation system by pinpointing the peptidase domain of F9, providing new insights for toxicological studies of novel pollutants.

This study explores the influence of after-school tutoring on reading comprehension skills of students with dyslexia (DD) in grades 3–5 in China and its participating factors. Using a mixed-methods design, the quantitative data (GORT-4 reading test) of 50 public primary school students in Qingdao and their parents’ feedback were collected through questionnaires, supplemented by semi-structured interviews with eight parents for qualitative analysis. The findings are as follows: (1) Family socioeconomic status and parents’ awareness of DD are the key factors to participate in after-school counseling; (2) The students who participated in the after-school counseling performed significantly better in the GORT-4 comprehension test (P < 0.05); (3) Counseling makes up for the lack of classroom learning through personalized strategies (such as multi-sensory teaching and phonological awareness training), but parents’ psychological support is the core of successful intervention. The necessity of early targeted intervention was emphasized, and practical enlightenment was provided for the educational support system of DD students in China.

Sleep efficiency is a key factor in human health. However, existing research tends to focus on isolated variables, ignoring the intricate relationship between sleep patterns and daily habits. In this paper, the Sleep Efficiency Dataset of the Kaggle platform was selected to analyze the effects of sleep structure and lifestyle factors on sleep efficiency. Through multiple linear regression models, the relationship between sleep duration, deep sleep percentage, light sleep percentage, rapid eye movement (REM) sleep percentage, and number of wake-ups was explored. The results showed that deep sleep and REM sleep percentage had a significant positive effect on sleep efficiency, while the number of wake-ups had a negative effect. The random forest model further analyzed the effect of lifestyle on sleep efficiency, and the results showed that age, alcohol consumption, and smoking status were the main factors affecting sleep efficiency. The results of the study provide data to support the improvement of sleep efficiency.

Dementia is a group of diseases that affect brain function and lead to a decline in cognitive abilities. Some types involve the gradual loss of function in brain neurons, such as Alzheimer's disease, which is characterized by progressive memory loss and cognitive dysfunction and is one of the most common neurodegenerative dementias. The growing prevalence, particularly in aging populations, underscores the need for enhanced diagnostic and therapeutic stproportiongies. The application of cutting-edge genomic technologies, encompassing genome-wide association studies (GWAS) to identify genetic variants associated with Alzheimer's disease (AD), whole exome sequencing (WES) to explore protein-coding regions for functional mutations, and single-cell RNA sequencing (scRNA-seq) to dissect cellular heterogeneity at a granular level, has collectively enhanced our insight into the intricate genetic architecture of AD, shedding light on the complex interplay of genes and biological pathways implicated in the disease's pathogenesis. GWAS has identified several risk loci, including the APOE ε4 allele, which plays a crucial role in lipid metabolism and amyloid-β clearance. However, GWAS primarily detects common variants, limiting its ability to explain rare genetic mutations. WES, which focuses on protein-coding regions, has uncovered rare variants in genes such as PSEN1 and PSEN2, which are linked to early-onset familial AD. Despite its strengths, WES overlooks non-coding regions critical for gene regulation. scRNA-seq provides high-resolution insights into cellular heterogeneity, identifying cell subsets and molecular mechanisms involved in AD. This technique has revealed novel therapeutic targets by analyzing gene expression at the single-cell level. However, it faces challenges in data processing and cost. Integrating these approaches offers a more comprehensive view of AD’s pathology, enhancing early diagnosis and personalized treatment. As genomic technologies evolve, they hold great promise for improving our understanding and management of AD.