The interactions between players of the prisoner's dilemma game are inferred using observed game data.All participants play the game with their counterparts and gain corresponding rewards during each round of the ...The interactions between players of the prisoner's dilemma game are inferred using observed game data.All participants play the game with their counterparts and gain corresponding rewards during each round of the game.The strategies of each player are updated asynchronously during the game.Two inference methods of the interactions between players are derived with naive mean-field(n MF)approximation and maximum log-likelihood estimation(MLE),respectively.Two methods are tested numerically also for fully connected asymmetric Sherrington-Kirkpatrick models,varying the data length,asymmetric degree,payoff,and system noise(coupling strength).We find that the mean square error of reconstruction for the MLE method is inversely proportional to the data length and typically half(benefit from the extra information of update times)of that by n MF.Both methods are robust to the asymmetric degree but work better for large payoffs.Compared with MLE,n MF is more sensitive to the strength of couplings and prefers weak couplings.展开更多
连接顺序选择是查询优化领域中极具挑战性的研究方向,对于数据库管理系统获得良好的查询性能至关重要.然而,传统优化方法和现有智能优化方法均存在着不足,如规划时间过长、容易得到质量较差的连接计划、编码未考虑结构特征、依赖基数估...连接顺序选择是查询优化领域中极具挑战性的研究方向,对于数据库管理系统获得良好的查询性能至关重要.然而,传统优化方法和现有智能优化方法均存在着不足,如规划时间过长、容易得到质量较差的连接计划、编码未考虑结构特征、依赖基数估计和代价估计使得连接计划无法反映真实的执行时间等.针对上述问题,提出了一种新型基于异步Dueling DQN(Deep Q-network)和计划时间预测网络的连接优化器:ADP-Join(Asynchronous Dueling DQN and Plan Latency Prediction Network for Join Order Selection).ADP-Join集成了一种新的编码方法,能够区分不同结构的连接计划.ADP-Join设计了计划时间预测网络PLN(Plan Latency Prediction Network)来改善现有基于强化学习优化器的奖励机制.再者,提出异步更新机制改进Dueling DQN模型来提升训练性能和减少训练时间.大量的实验结果表明,在TPC-H和JOB真实数据集上ADP-Join的性能优于现有的智能优化器.展开更多
为改进传统RAID5机制小写性能低下问题,提出了一种高性能的冗余数据读写机制RMB(RedundancyManagement for RAID5).RMB采用异步冗余校验方法APC(Asynchronous Parity Calculating),对现有RAID5更新算法的关键步骤进行了优化,将统一计算...为改进传统RAID5机制小写性能低下问题,提出了一种高性能的冗余数据读写机制RMB(RedundancyManagement for RAID5).RMB采用异步冗余校验方法APC(Asynchronous Parity Calculating),对现有RAID5更新算法的关键步骤进行了优化,将统一计算校验块的同步操作拆分成同步镜像写更新和异步冗余计算更新.该操作分解了数据与其校验数据更新的紧耦合现状,提升了写操作并发度;通过合并多个版本镜像块写操作减少异或计算操作的次数.通过实验验证表,RMB机制,可将小写性能大幅提高,峰值可达4倍以上.展开更多
基金supported by the National Natural Science Foundation of China(Grant Nos.11705079 and 11705279)the Scientific Research Foundation of Nanjing University of Posts and Telecommunications(Grant Nos.NY221101 and NY222134)the Science and Technology Innovation Training Program(Grant No.STITP 202210293044Z)。
文摘The interactions between players of the prisoner's dilemma game are inferred using observed game data.All participants play the game with their counterparts and gain corresponding rewards during each round of the game.The strategies of each player are updated asynchronously during the game.Two inference methods of the interactions between players are derived with naive mean-field(n MF)approximation and maximum log-likelihood estimation(MLE),respectively.Two methods are tested numerically also for fully connected asymmetric Sherrington-Kirkpatrick models,varying the data length,asymmetric degree,payoff,and system noise(coupling strength).We find that the mean square error of reconstruction for the MLE method is inversely proportional to the data length and typically half(benefit from the extra information of update times)of that by n MF.Both methods are robust to the asymmetric degree but work better for large payoffs.Compared with MLE,n MF is more sensitive to the strength of couplings and prefers weak couplings.
文摘连接顺序选择是查询优化领域中极具挑战性的研究方向,对于数据库管理系统获得良好的查询性能至关重要.然而,传统优化方法和现有智能优化方法均存在着不足,如规划时间过长、容易得到质量较差的连接计划、编码未考虑结构特征、依赖基数估计和代价估计使得连接计划无法反映真实的执行时间等.针对上述问题,提出了一种新型基于异步Dueling DQN(Deep Q-network)和计划时间预测网络的连接优化器:ADP-Join(Asynchronous Dueling DQN and Plan Latency Prediction Network for Join Order Selection).ADP-Join集成了一种新的编码方法,能够区分不同结构的连接计划.ADP-Join设计了计划时间预测网络PLN(Plan Latency Prediction Network)来改善现有基于强化学习优化器的奖励机制.再者,提出异步更新机制改进Dueling DQN模型来提升训练性能和减少训练时间.大量的实验结果表明,在TPC-H和JOB真实数据集上ADP-Join的性能优于现有的智能优化器.
文摘为改进传统RAID5机制小写性能低下问题,提出了一种高性能的冗余数据读写机制RMB(RedundancyManagement for RAID5).RMB采用异步冗余校验方法APC(Asynchronous Parity Calculating),对现有RAID5更新算法的关键步骤进行了优化,将统一计算校验块的同步操作拆分成同步镜像写更新和异步冗余计算更新.该操作分解了数据与其校验数据更新的紧耦合现状,提升了写操作并发度;通过合并多个版本镜像块写操作减少异或计算操作的次数.通过实验验证表,RMB机制,可将小写性能大幅提高,峰值可达4倍以上.
