仰恩大学

Rotary Position Embedding (RoPE) enables Transformer models to extrapolate beyond their training sequence lengths, yet the resulting generation quality remains under-characterized. We present a systematic empirical study of RoPE-based extrapolation using multiple complementary metrics. Using a 10.4M-parameter GPT-style model trained on 256-token sequences (TinyShakespeare), we evaluate generation at 2x and 4x lengths (512 and 1024 tokens) with perplexity, n-gram repetition, and a heuristic coherence proxy. We observe a pronounced divergence among metrics: self-perplexity can even improve at 4x while repetition rises sharply ( 6x increase in trigram repetition from baseline to 4x ) and coherence remains low and unstable at 4x (as low as 0.83%). A supplemental validation-set PPL analysis shows monotonic degradation with length, confirming that self-PPL can be misleading under degenerative loops. This indicates perplexity alone is unreliable for judging long-range generation quality; models may retain local statistical fit while losing global discourse structure. We additionally report robustness across seeds, PE baselines (ALiBi and sinusoidal), dataset diversity (WikiText-103), decoding strategy comparisons, scale sensitivity across 1.3B and 7B models, and qualitative samples. We contribute: (1) a multi-metric evaluation framework for extrapolation quality, (2) empirical evidence of metric divergence under extrapolation, and (3) practical guidelines suggesting modest extrapolation may be usable, whereas extreme extrapolation requires careful quality safeguards.

This research comprehensively examined regional variations in individualism and collectivism within China and the explanatory power of 20 potential factors derived from the existing cultural psychological theories, including socioecological (e.g., climates, subsistence activities, and modernization) and population (e.g., ethnicity, genetic and linguistic features) ones. Three large-scale studies were conducted. In Study 1 ( N = 18,606), we assessed 11 markers of individualism-collectivism. In Studies 2 and 3, we evaluated whether the main findings of Study 1 could be replicated in two large community samples ( N = 8,167 and 20,590, respectively) when perceived cultures of individualism and collectivism were considered and when the pandemic of COVID-19 was going on, respectively. Three studies consistently showed that (1) systematic East-West differences in individualism and collectivism existed, with higher individualism and lower collectivism in Eastern China than in Western China; meanwhile, there was some suggestive evidence of North-South differences, with higher individualism and lower collectivism in Southern China than in Northern China; (2) among the predictive socioecological (e.g., pathogen prevalence, modernization) and population factors (e.g., ethnicity, genetic and linguistic relatedness), modernization was the most potent and robust predictor of individualism and collectivism; (3) regional cultural variations often arose from the combined influences of multiple factors; and (4) the evidence supported most existing theories about the sources of cultural variation, with a few exceptions (e.g., the rice theory). These findings provide novel insights into regional cultural variations and the applicability of many relevant theories within China, and also shed light on cross-cultural psychology in general.

This study examines the role of academic independent directors in shaping corporate environmental performance, thereby addressing gaps in the existing literature. Using panel data (2009-2022) from Chinese listed firms, we find that academic independent directors significantly improve corporate environmental performance. This improvement is primarily driven by enhanced environmental awareness, reduced managerial myopia, and increased green innovation. The effects vary across stakeholder capabilities, managerial green awareness, and the professional backgrounds of academic independent directors. These findings provide new insights into the relationship between academic independent directors and corporate environmental performance, and offer important implications for strengthening their role in corporate governance.

The role of financial institutions in climate governance is increasingly being recognized, particularly in relation to Scope 3 emissions. While existing research has focused primarily on lending and investment activities, the potential influence of insurance operations on lifecycle emissions remains underexplored. This study examines electric vehicle (EV) insurance underwriting as a form of indirect climate governance, with particular attention being paid to claim-related decision processes that affect repair-, replacement-, and battery-related outcomes. A decision-analytical, scenario-based portfolio model is developed to analyze how underwriting and claims parameters may influence lifecycle emissions exposure. The model incorporates literature-informed and scenario-based parameter ranges derived from the lifecycle assessment literature and industry-relevant assumptions, while explicitly accounting for regulatory, technical, and behavioral constraints that limit insurer decision making. An exposure-based attribution framework is applied to link insurance-mediated outcomes to emissions associated with vehicle and battery manufacturing. The results suggest that claim-related parameters—particularly total-loss probability—are associated with variations in modeled emissions exposure within the analytical framework. Scenario analysis indicates that, under plausible parameter configurations, differences in claims decision structures may contribute to variation in lifecycle emissions at the portfolio level. Sensitivity analysis further indicates that these relationships appear stable across a range of parameter assumptions. The findings should be interpreted as scenario-based insights rather than empirical estimates, highlighting potential pathways through which insurance operations may influence emissions outcomes within existing constraints. The study contributes to the literature by extending Scope 3 governance analysis to insurance and by proposing an operational framework for interpreting insurance-associated emissions in lifecycle terms.

The rapid diffusion of electric vehicles (EVs) is reshaping mobility markets and creating new opportunities for embedded financial services. This study examines consumer acceptance of embedded EV insurance, which refers to coverage bundled directly at the point of vehicle sale in Taiwan, China. Using survey data from 400 licensed drivers, we analyze how demographic factors and five psychological drivers—perceived savings, convenience, trust, expected satisfaction, and fairness—shape the likelihood of choosing embedded insurance over traditional stand-alone policies. Welch’s t-tests show that younger drivers perceive greater savings and convenience, while older drivers express stronger fairness concerns. Logistic regression results indicate that convenience (OR = 2.05) and perceived savings (OR = 1.76) substantially increase adoption likelihood, whereas fairness concerns reduce it (OR = 0.71). Theoretically, this study advances consumer behavior research by demonstrating how functional value perceptions (convenience and savings) and fairness evaluations jointly influence decisions in digitally mediated insurance contexts. It also contributes to embedded finance theory by revealing how insurance embedded within EV purchasing ecosystems reshapes consumer decision processes and alters traditional insurer–consumer relationships. These findings offer strategic implications for automakers, insurers, and policymakers designing consumer-centric embedded financial products in emerging mobility markets.