基于样本冗余度的主动学习优化方法

主动学习解决了因训练样本过大而导致需要大量人力物力的问题,核心问题是如何选择有价值的样本减少标注成本.以神经网络为分类器,大多数方法选择信息量大的样本并没有考虑所选择样本间的信息冗余问题.通过对冗余问题的研究,提出一种降低信息冗余的样本选择优化方法.用不确定性方法选出信息量大的样本构成候选样本集,同时用网络中计算的潜变...

ACTIVE LEARNING OPTIMIZATION METHOD BASED ON SAMPLE REDUNDANCY

Active learning solves the problem of excessive training samples,and the core problem is sample selection.With neural networks as classifiers,most methods choose a large amount of information and do not consider the information redundancy between selected samples.By researching the redundancy problem,a sample selection optimization method for reducing information redundancy is proposed.This method used the uncertainty method to select the sample with large amount of information to form the candidate sample set,and the vector of latent variables calculated in the network was represented.The sample information used the vector to calculate the cosine distance between the candidate samples to select a subset with a large separation distance and low information redundancy.Compared with several uncertainty methods in the mnist,fashion-mnist,and cifar-10 data sets,the lowest sample size can be reduced by 11%with the same sample accuracy.