一种基于少样本数据的在线主动学习与分类方法

杨静, 赵文仓, 徐越, 冯旸赫, 黄金才

An on⁃line active learning and classification method based on small sample data

Jing Yang, Wencang Zhao, Yue Xu, Yanghe Feng, Jincai Huang

表4 MNIST,CIFAR?10,CIFAR?100在不同训练集上的时间耗费比较

Table 4 Comparison for time consuming by all data and online fine?tune data for MNIST,CIFAR?10,CIFAR?100

方法 MNIST CIFAR?10 CIFAR?100 100轮耗时(s) 每轮平均耗时(s) 100轮耗时(s) 每轮平均耗时(s) 100轮耗时(s) 每轮平均耗时(s) 变异比?AD 33699.99 336.99 520540.78 5205.40 833933.65 8339.33 变异比?OD 18809.54 188.09 21807.34 218.07 34354.54 343.54 预测熵?AD 30081.62 300.81 487761.18 4877.61 813504.12 8135.04 预测熵?OD 17943.67 179.43 21043.45 210.43 34065.86 340.65 BALD?AD 29989.71 299.89 484232.12 4842.32 803423.55 8034.23 BALD?OD 17353.88 173.53 20942.65 209.42 33901.54 339.01 CIS?AD 49023.34 490.23 689832.73 6898.32 1232323.51 12323.23 CIS?OD 21302.62 213.02 30231.51 302.31 39515.21 395.15