针对大数据环境下DCNN(deep convolutional neural network)算法中存在网络冗余参数过多、参数寻优能力不佳和并行效率低的问题,提出了大数据环境下基于特征图和并行计算熵的深度卷积神经网络算法MR-FPDCNN(deep convolutional neural n...针对大数据环境下DCNN(deep convolutional neural network)算法中存在网络冗余参数过多、参数寻优能力不佳和并行效率低的问题,提出了大数据环境下基于特征图和并行计算熵的深度卷积神经网络算法MR-FPDCNN(deep convolutional neural network algorithm based on feature graph and parallel computing entropy using MapReduce)。该算法设计了基于泰勒损失的特征图剪枝策略FMPTL(feature map pruning based on Taylor loss),预训练网络,获得压缩后的DCNN,有效减少了冗余参数,降低了DCNN训练的计算代价。提出了基于信息共享搜索策略ISS(information sharing strategy)的萤火虫优化算法IFAS(improved firefly algorithm based on ISS),根据“IFAS”算法初始化DCNN参数,实现DCNN的并行化训练,提高网络的寻优能力。在Reduce阶段提出了基于并行计算熵的动态负载均衡策略DLBPCE(dynamic load balancing strategy based on parallel computing entropy),获取全局训练结果,实现了数据的快速均匀分组,从而提高了集群的并行效率。实验结果表明,该算法不仅降低了DCNN在大数据环境下训练的计算代价,而且提高了并行系统的并行化性能。展开更多
Turbo decoding is iterative decoding, and the MAP algorithm isoptimal in terms of performance in Turbo decoding. The log-MAPalgorithms is the MAP executed in the logarithmic domain, so it isalso optimal. Both the MAP ...Turbo decoding is iterative decoding, and the MAP algorithm isoptimal in terms of performance in Turbo decoding. The log-MAPalgorithms is the MAP executed in the logarithmic domain, so it isalso optimal. Both the MAP and the log-MAP algorithm are complicatedfor implementation. The max-log MAP algorithm is de- Rived from thelog-MAP with approximation, which is simply compared with the log-MAPalgorithm but is subopti- Malin terms of performance. A modifiedmax-log-MAP algorithm is presented in this paper, based on the TaylorSeries of logarithm and exponent. Analysis and simulation resultsshow that modified max-log-MAP algorithm Outperforms the max-log-MAPalgorithm with almost the same complexity.展开更多
In this paper, the flow patterns observed in horizontal Couette-Taylor flow（CTF） were correlated using dimensionless numbers. The analysis of the results showed that the structure of the flow was an outcome of inter...In this paper, the flow patterns observed in horizontal Couette-Taylor flow（CTF） were correlated using dimensionless numbers. The analysis of the results showed that the structure of the flow was an outcome of interaction between fluid inertia related to axial and rotational flows and gravitation. Therefore, the flow structures were correlated using axial and angular Reynolds numbers, and Archimedes number for the given value of gas-to-liquid flow ratio. Finally, the correlation for the prediction of the transition to the flow regime observed at high rotational speeds was proposed. The comparison with experiments carried out in the vertical CTF from the literature showed that this correlation can also be useful in the case of vertical flow.展开更多
文摘针对大数据环境下DCNN(deep convolutional neural network)算法中存在网络冗余参数过多、参数寻优能力不佳和并行效率低的问题,提出了大数据环境下基于特征图和并行计算熵的深度卷积神经网络算法MR-FPDCNN(deep convolutional neural network algorithm based on feature graph and parallel computing entropy using MapReduce)。该算法设计了基于泰勒损失的特征图剪枝策略FMPTL(feature map pruning based on Taylor loss),预训练网络,获得压缩后的DCNN,有效减少了冗余参数,降低了DCNN训练的计算代价。提出了基于信息共享搜索策略ISS(information sharing strategy)的萤火虫优化算法IFAS(improved firefly algorithm based on ISS),根据“IFAS”算法初始化DCNN参数,实现DCNN的并行化训练,提高网络的寻优能力。在Reduce阶段提出了基于并行计算熵的动态负载均衡策略DLBPCE(dynamic load balancing strategy based on parallel computing entropy),获取全局训练结果,实现了数据的快速均匀分组,从而提高了集群的并行效率。实验结果表明,该算法不仅降低了DCNN在大数据环境下训练的计算代价,而且提高了并行系统的并行化性能。
文摘Turbo decoding is iterative decoding, and the MAP algorithm isoptimal in terms of performance in Turbo decoding. The log-MAPalgorithms is the MAP executed in the logarithmic domain, so it isalso optimal. Both the MAP and the log-MAP algorithm are complicatedfor implementation. The max-log MAP algorithm is de- Rived from thelog-MAP with approximation, which is simply compared with the log-MAPalgorithm but is subopti- Malin terms of performance. A modifiedmax-log-MAP algorithm is presented in this paper, based on the TaylorSeries of logarithm and exponent. Analysis and simulation resultsshow that modified max-log-MAP algorithm Outperforms the max-log-MAPalgorithm with almost the same complexity.
文摘In this paper, the flow patterns observed in horizontal Couette-Taylor flow（CTF） were correlated using dimensionless numbers. The analysis of the results showed that the structure of the flow was an outcome of interaction between fluid inertia related to axial and rotational flows and gravitation. Therefore, the flow structures were correlated using axial and angular Reynolds numbers, and Archimedes number for the given value of gas-to-liquid flow ratio. Finally, the correlation for the prediction of the transition to the flow regime observed at high rotational speeds was proposed. The comparison with experiments carried out in the vertical CTF from the literature showed that this correlation can also be useful in the case of vertical flow.
