Non-Linear Dimensionality Reduction using Auto-Encoder for Optimized Malaria Infected Blood Cell Classifier
Date of Award
Computer and Information Science
Neural Networks have been widely used in the problem of Medical Image Analysis. However, when dealing with large images, deep networks easily exhaust computer resources, which in turn hinders training. This paper shows the efficacy of using Auto-Encoders as a dimensionality reduction tool to increase the efficiency of a Malaria Infected Blood Cell Image classifier. We show that using an autoencoder, we can reduce the dimensionality of large blood cell images effectively such that the features in the new space retain all the essential information from the original input. Then we show that the new features obtained from the autoencoder can be used to train a classifier while maintaining the same accuracy. Using a Convolutional autoencoder with a Convolutional Neural Network(CNN) for malaria infected blood cell classification, gives us a significantly smaller model compared to a vanilla CNN model which performs similar in terms of accuracy.
Dhakal, Aayush, "Non-Linear Dimensionality Reduction using Auto-Encoder for Optimized Malaria Infected Blood Cell Classifier" (2021). Honors Theses. 1873.