共识机制是区块链技术的重要组成部分,但是主流的共识机制尤其是工作量证明共识机制都存在算力过度耗费和吞吐量低等问题.而联邦学习作为一种分布式机器学习方法,学习模型的本地训练和最终的参与方贡献度计算都需要消耗大量算力资源.因...共识机制是区块链技术的重要组成部分,但是主流的共识机制尤其是工作量证明共识机制都存在算力过度耗费和吞吐量低等问题.而联邦学习作为一种分布式机器学习方法,学习模型的本地训练和最终的参与方贡献度计算都需要消耗大量算力资源.因此,提出了一种支持自适应联邦学习任务的可信公平区块链框架TFchain,探索如何利用原本共识机制中耗费的大量算力来提高联邦学习的效率.首先,设计了基于区块链和联邦学习的全新共识机制PoTF(proof of trust and fair),该共识机制将区块链的节点设置为联邦学习的参与方,将原本共识机制中用于哈希计算的大量无效算力转移到联邦学习中,进行本地模型的训练和参与方贡献度的评估;其次,在提高区块链交易吞吐量的同时,对联邦学习的参与方进行了合理的贡献度评估和激励;最后,设计了防止节点作恶的算法.实验结果表明,提出的TFchain能够在回收算力的同时有效提升区块链的交易处理性能,对积极参与联邦学习的参与方进行有效正向的激励.展开更多
The group-contribution (GC) methods suffer from a limitation concerning to the prediction of process-related indexes, e.g., thermal efficiency. Recently developed analytical models for thermal efficiency of organic Ra...The group-contribution (GC) methods suffer from a limitation concerning to the prediction of process-related indexes, e.g., thermal efficiency. Recently developed analytical models for thermal efficiency of organic Rankine cycles (ORCs) provide a possibility of overcoming the limitation of the GC methods because these models formulate thermal efficiency as functions of key thermal properties. Using these analytical relations together with GC methods, more than 60 organic fluids are screened for medium-low temperature ORCs. The results indicate that the GC methods can estimate thermal properties with acceptable accuracy (mean relative errors are 4.45%-11.50%);the precision, however, is low because the relative errors can vary from less than 0.1% to 45.0%. By contrast, the GC-based estimation of thermal efficiency has better accuracy and precision. The relative errors in thermal efficiency have an arithmetic mean of about 2.9% and fall within the range of 0-24.0%. These findings suggest that the analytical equations provide not only a direct way of estimating thermal efficiency but an accurate and precise approach to evaluating working fluids and guiding computer-aided molecular design of new fluids for ORCs using GC methods.展开更多
文摘共识机制是区块链技术的重要组成部分,但是主流的共识机制尤其是工作量证明共识机制都存在算力过度耗费和吞吐量低等问题.而联邦学习作为一种分布式机器学习方法,学习模型的本地训练和最终的参与方贡献度计算都需要消耗大量算力资源.因此,提出了一种支持自适应联邦学习任务的可信公平区块链框架TFchain,探索如何利用原本共识机制中耗费的大量算力来提高联邦学习的效率.首先,设计了基于区块链和联邦学习的全新共识机制PoTF(proof of trust and fair),该共识机制将区块链的节点设置为联邦学习的参与方,将原本共识机制中用于哈希计算的大量无效算力转移到联邦学习中,进行本地模型的训练和参与方贡献度的评估;其次,在提高区块链交易吞吐量的同时,对联邦学习的参与方进行了合理的贡献度评估和激励;最后,设计了防止节点作恶的算法.实验结果表明,提出的TFchain能够在回收算力的同时有效提升区块链的交易处理性能,对积极参与联邦学习的参与方进行有效正向的激励.
基金Project(51778626) supported by the National Natural Science Foundation of China
文摘The group-contribution (GC) methods suffer from a limitation concerning to the prediction of process-related indexes, e.g., thermal efficiency. Recently developed analytical models for thermal efficiency of organic Rankine cycles (ORCs) provide a possibility of overcoming the limitation of the GC methods because these models formulate thermal efficiency as functions of key thermal properties. Using these analytical relations together with GC methods, more than 60 organic fluids are screened for medium-low temperature ORCs. The results indicate that the GC methods can estimate thermal properties with acceptable accuracy (mean relative errors are 4.45%-11.50%);the precision, however, is low because the relative errors can vary from less than 0.1% to 45.0%. By contrast, the GC-based estimation of thermal efficiency has better accuracy and precision. The relative errors in thermal efficiency have an arithmetic mean of about 2.9% and fall within the range of 0-24.0%. These findings suggest that the analytical equations provide not only a direct way of estimating thermal efficiency but an accurate and precise approach to evaluating working fluids and guiding computer-aided molecular design of new fluids for ORCs using GC methods.
