The analysis of the ejaculate,better known as spermiogram,represents the first and main step to identify whether a series of sperm quality parameters are within the norm and therefore are consistent with normal sperm ...The analysis of the ejaculate,better known as spermiogram,represents the first and main step to identify whether a series of sperm quality parameters are within the norm and therefore are consistent with normal sperm fertilizing capacity.Among these,sperm concentration and motility are the first parameters to be evaluated through an estimation carried out by expert examiners.展开更多
Abstract When interfacial layers are viewed as a separate phase, the interface thickness plays an essential role in assessing physico-mechanical properties of particulate materials. However, the interface thickness fr...Abstract When interfacial layers are viewed as a separate phase, the interface thickness plays an essential role in assessing physico-mechanical properties of particulate materials. However, the interface thickness from sectional analysis is often overestimated, due to the irregularity of surface textures of grains in opaque materials that gives rise to the normal of a cross-sectional plane non-perpendicular to the surface of grains. Hence, the determination of the overestimation degree is very critical to precisely obtain the interface thickness. This article develops a numerical model for the overestimation degree of the interface thickness around an ellipsoidal grain with an arbitrary aspect ratio, by applying an accurate sectional analysis algorithm, and quantitative stereology and geometrical probability theories. Furthermore, on the basis of the developed numerical model, the influence of ellipsoidal particle shape on the overestimation degree is quantitatively charac-terized.展开更多
离线强化学习旨在仅通过使用预先收集的离线数据集进行策略的有效学习,从而减少与环境直接交互所带来的高昂成本。然而,由于缺少环境对智能体行为的交互反馈,从离线数据集中学习到的策略可能会遇到数据分布偏移的问题,进而导致外推误差...离线强化学习旨在仅通过使用预先收集的离线数据集进行策略的有效学习,从而减少与环境直接交互所带来的高昂成本。然而,由于缺少环境对智能体行为的交互反馈,从离线数据集中学习到的策略可能会遇到数据分布偏移的问题,进而导致外推误差的不断加剧。当前方法多采用策略约束或模仿学习方法来缓解这一问题,但其学习到的策略通常较为保守。针对上述难题,提出一种基于自适应分位数的方法。具体而言,该方法在双Q估计的基础上进一步利用双Q的估计差值大小对分布外未知动作的价值高估情况进行评估,同时结合分位数思想自适应调整分位数来校正过估计偏差。此外,构建分位数优势函数作为策略约束项权重以平衡智能体对数据集的探索和模仿,从而缓解策略学习的保守性。最后在D4RL(datasets for deep data-driven reinforcement learning)数据集上验证算法的有效性,该算法在多个任务数据集上表现优异,同时展现出在不同场景应用下的广泛潜力。展开更多
文摘The analysis of the ejaculate,better known as spermiogram,represents the first and main step to identify whether a series of sperm quality parameters are within the norm and therefore are consistent with normal sperm fertilizing capacity.Among these,sperm concentration and motility are the first parameters to be evaluated through an estimation carried out by expert examiners.
基金supported by the Natural Science Foundation Project for Jiangsu Province(BK20130841)National Science Foundation Project for Distinguished Young Scholars(11125208)the Ministry of Science and Technology of China(973 Project)(2009CB623203 and 2010CB832702)
文摘Abstract When interfacial layers are viewed as a separate phase, the interface thickness plays an essential role in assessing physico-mechanical properties of particulate materials. However, the interface thickness from sectional analysis is often overestimated, due to the irregularity of surface textures of grains in opaque materials that gives rise to the normal of a cross-sectional plane non-perpendicular to the surface of grains. Hence, the determination of the overestimation degree is very critical to precisely obtain the interface thickness. This article develops a numerical model for the overestimation degree of the interface thickness around an ellipsoidal grain with an arbitrary aspect ratio, by applying an accurate sectional analysis algorithm, and quantitative stereology and geometrical probability theories. Furthermore, on the basis of the developed numerical model, the influence of ellipsoidal particle shape on the overestimation degree is quantitatively charac-terized.
文摘离线强化学习旨在仅通过使用预先收集的离线数据集进行策略的有效学习,从而减少与环境直接交互所带来的高昂成本。然而,由于缺少环境对智能体行为的交互反馈,从离线数据集中学习到的策略可能会遇到数据分布偏移的问题,进而导致外推误差的不断加剧。当前方法多采用策略约束或模仿学习方法来缓解这一问题,但其学习到的策略通常较为保守。针对上述难题,提出一种基于自适应分位数的方法。具体而言,该方法在双Q估计的基础上进一步利用双Q的估计差值大小对分布外未知动作的价值高估情况进行评估,同时结合分位数思想自适应调整分位数来校正过估计偏差。此外,构建分位数优势函数作为策略约束项权重以平衡智能体对数据集的探索和模仿,从而缓解策略学习的保守性。最后在D4RL(datasets for deep data-driven reinforcement learning)数据集上验证算法的有效性,该算法在多个任务数据集上表现优异,同时展现出在不同场景应用下的广泛潜力。
