75 research outputs found
Enhancing the Hierarchical Environment Design via Generative Trajectory Modeling
No full textUnsupervised Environment Design (UED) is a paradigm for automatically
generating a curriculum of training environments, enabling agents trained in
these environments to develop general capabilities, i.e., achieving good
zero-shot transfer performance. However, existing UED approaches focus
primarily on the random generation of environments for open-ended agent
training. This is impractical in scenarios with limited resources, such as the
constraints on the number of generated environments. In this paper, we
introduce a hierarchical MDP framework for environment design under resource
constraints. It consists of an upper-level RL teacher agent that generates
suitable training environments for a lower-level student agent. The RL teacher
can leverage previously discovered environment structures and generate
environments at the frontier of the student's capabilities by observing the
student policy's representation. Moreover, to reduce the time-consuming
collection of experiences for the upper-level teacher, we utilize recent
advances in generative modeling to synthesize a trajectory dataset to train the
teacher agent. Our proposed method significantly reduces the resource-intensive
interactions between agents and environments and empirical experiments across
various domains demonstrate the effectiveness of our approach
Diversity Induced Environment Design via Self-Play
No full textRecent work on designing an appropriate distribution of environments has
shown promise for training effective generally capable agents. Its success is
partly because of a form of adaptive curriculum learning that generates
environment instances (or levels) at the frontier of the agent's capabilities.
However, such an environment design framework often struggles to find effective
levels in challenging design spaces and requires costly interactions with the
environment. In this paper, we aim to introduce diversity in the Unsupervised
Environment Design (UED) framework. Specifically, we propose a task-agnostic
method to identify observed/hidden states that are representative of a given
level. The outcome of this method is then utilized to characterize the
diversity between two levels, which as we show can be crucial to effective
performance. In addition, to improve sampling efficiency, we incorporate the
self-play technique that allows the environment generator to automatically
generate environments that are of great benefit to the training agent.
Quantitatively, our approach, Diversity-induced Environment Design via
Self-Play (DivSP), shows compelling performance over existing methods
Regional Differences, Distributional Dynamics and Spatial Convergence of Pharmacist Human Resources in China: A Healthcare Tiers Difference Perspective
No full textBo Ding,1 Run Zhao,2 Xinyi Yue,2 Dexun Li2 1Pharmacy Department, Hefei Maternal and Child Health Hospital, Hefei, People’s Republic of China; 2School of Pharmaceutical Economics and Management, Anhui University of Chinese Medicine, Hefei, People’s Republic of ChinaCorrespondence: Dexun Li, Email [email protected]: This study aims to analyze the regional and hierarchical disparities in the distribution of pharmacist human resources in China from 2012 to 2022, with the goal of identifying underlying trends and challenges. The findings are intended to serve as a basis for optimizing the equitable and efficient allocation of pharmacist resources to better support healthcare system development.Patients and Methods: Data were collected from the China Health Statistics Yearbook and China Statistics Yearbook. The number of pharmacists per 1,000 population was selected as a measure of equity in pharmacist allocation, and Dagum’s Gini coefficient, kernel density estimation method and spatial β-convergence model were utilized to analyze the regional differences and convergence trends of pharmacist resource allocation in Chinese hospitals and primary health care organizations.Results: The findings reveal that while the overall allocation of pharmacist resources has improved, significant discrepancies remain. The mean number of hospital pharmacists was approximately twice that of primary care centers, with the most pronounced disparity observed in the central region. Regional disparities show declining trends in hospitals but widening gaps in PHCs, particularly in the eastern region. Kernel density results highlight improvements in pharmacist distribution but also reinforce advantages for High-resource provinces, especially at the hospital tier. Spatial analysis indicates significant clustering effects in pharmacist allocation, though these effects have weakened over time. Notably, absolute and conditional β-convergence trends are observed, with faster convergence rates in hospitals compared to PHCs and distinct regional variations in convergence speed.Conclusion: Since 2012, the allocation of pharmacists’ human resources in China has improved, with regional disparities showing signs of reduction. However, hierarchical disparities remain a significant issue that requires further attention, particularly in the central region. To address these challenges, it is essential to increase investment in primary healthcare institutions, with a focus on strengthening pharmacist staffing, improving infrastructure, and enhancing the capacity of pharmacy services at the grassroots level.Keywords: pharmacists, health resource allocation, primary healthcare institutions, regional disparities, β convergenc
- …
