Application of ResNet-50, Convolutional Neural Network and Multi-class Logistic Regression in Skin Cancer Classification and Analysis of Back Propagation
In this project we aim at using two neural network architectures to classify skin cancer: Residual Network(ResNet) and a traditional Convolutional Neural Network (CNN). We experimentally observed and compared the performance between these two models. Furthermore, we understood the mathematics behind the architecture of ResNet, back-propagation algorithm and Adam optimization. The experimental results show that traditional CNN has a slightly better performance than ResNet, but due to the restrictions of the computational power, we cannot train ResNet with a large amount of epochs, because it is a deep network. 1