基于混合核平滑自编码器的分类器设计

摘要

Autoencoder is a kind of neural network that reconstructs input data and has strong learning ability which has been widely applied to many kinds of classification tasks.In order to apply autoencoder to classification tasks,this paper proposes a method based on hybrid kernel smooth autoencoder to classify handwritten numerals and face images.We compared the weights of different kernel functions,and compared the results with sparse self-encoders.The experimental results show that hybrid kernel based smooth autoencoder in a certain ratio can obtain better classification results than the single kernel function autoencoder.%自编码器是一种重构输入数据的神经网络,具有较强的特征学习能力,被广泛地应用在各种分类任务中.为了将自编码器应用于分类任务中,研究一种基于混合核平滑自编码器的方法进行手写数字和人脸图像的分类.对不同核函数的权重进行比较,同时还将结果与稀疏自编码器进行比较.实验结果表明基于混合核的平滑自编码在一定的比例下可以获得比单个核函数更好的分类效果.