This project is a multi-model deep learning implementation on the CIFAR10 dataset, which is a collection of 10 categories of small natural images such as animals, vehicles, etc. The project implements three different models using TensorFlow, Keras and Numpy. The models include a simple dense neural network, a dense neural network with batch normalization, and a convolutional neural network.
The project uses the Keras library to build the models and uses the CIFAR10 dataset to train and evaluate the models. The training process uses the Stochastic Gradient Descent (SGD) optimizer and the categorical cross-entropy loss function. The project also includes the use of EarlyStopping callback to stop the training process when the model performance reaches a saturation point.
The project provides a comprehensive understanding of how to build, train, and evaluate different deep learning models for image classification tasks. The project demonstrates the impact of different architectural decisions such as adding batch normalization, increasing the number of layers, and using convolutional layers on the overall model performance.