为了提高深度卷积神经网络(DCNN)的图像并行处理能力,提高其图像识别的准确率和运行效率,研究过程以MapReduce并行计算框架和从图像到矩阵(Image to Column,Im2col)算法,分别进行原始图像特征并行提取和筛选、模型并行训练和参数并行更...为了提高深度卷积神经网络(DCNN)的图像并行处理能力,提高其图像识别的准确率和运行效率,研究过程以MapReduce并行计算框架和从图像到矩阵(Image to Column,Im2col)算法,分别进行原始图像特征并行提取和筛选、模型并行训练和参数并行更新,构建了并行DCNN优化算法。在性能检测阶段,将全连接神经网络和基于特征图和并行计算熵的深度卷积神经网络算法作为对照组,对比TOP⁃1准确率、浮点运算量、损失函数振荡性、运算时长四项指标,结果显示,此次提出的并行DCNN优化算法性能最佳。展开更多
Understanding and managing charge carrier recombination dynamics is crucial for optimizing the performance of metal halide perovskite optoelectronic devices.In this work,we introduce a machine learning-assisted intens...Understanding and managing charge carrier recombination dynamics is crucial for optimizing the performance of metal halide perovskite optoelectronic devices.In this work,we introduce a machine learning-assisted intensity-modulated two-photon photoluminescence microscopy approach for quantitatively mapping recombination processes in MAPbBr_(3) perovskite microcrystalline films at micrometer-scale resolution.To enhance model accuracy,a balanced classification sampling strategy was applied during the machine learning optimization stage.展开更多
文摘为了提高深度卷积神经网络(DCNN)的图像并行处理能力,提高其图像识别的准确率和运行效率,研究过程以MapReduce并行计算框架和从图像到矩阵(Image to Column,Im2col)算法,分别进行原始图像特征并行提取和筛选、模型并行训练和参数并行更新,构建了并行DCNN优化算法。在性能检测阶段,将全连接神经网络和基于特征图和并行计算熵的深度卷积神经网络算法作为对照组,对比TOP⁃1准确率、浮点运算量、损失函数振荡性、运算时长四项指标,结果显示,此次提出的并行DCNN优化算法性能最佳。
基金financial support from Swedish Energy Agency grant 50709-1Swedish Research Council grant 2021-05207+2 种基金KAW WISE/WASP grant 01-22Olle Engkvist foundation grant 235-0422Light and Materials profile area at Lund University(Young Investigator Synergy Award,2023)。
文摘Understanding and managing charge carrier recombination dynamics is crucial for optimizing the performance of metal halide perovskite optoelectronic devices.In this work,we introduce a machine learning-assisted intensity-modulated two-photon photoluminescence microscopy approach for quantitatively mapping recombination processes in MAPbBr_(3) perovskite microcrystalline films at micrometer-scale resolution.To enhance model accuracy,a balanced classification sampling strategy was applied during the machine learning optimization stage.
