The project was created during my studies in Ireland in the Smart Technologies module, where we have to build a classification model using CIFAR-10 and CIFAR-100 datasets. A report was written under the file name "ca1_report.pdf".
The primary objective of the project is to build a convolutional neural network model that can classify between twenty-four different classes of the CIFAR-10 and CIFAR-100 datasets. The classes are car, bird, cat, deer, dog, horse, truck, cattle, fox, baby, boy, girl, man, woman, rabbit, squirrel, tree, bicycle, bus, motorcycle, pickup truck, train, lawn mower and tractor. This involves obtaining, preparing and exploring the data sets. Several CNN models are then constructed and fine-tuned to achieve optimal classification performance. Hyperparameter tuning is guided by continuous evaluation of model accuracy. Our focus is not only on achieving high accuracy, but also on improving the fairness of model predictions, especially across classes with unbalanced representation. The aim of this project is to present the CNN model that achieves the highest test accuracy, demonstrating its ability to accurately classify images across the diverse set of twenty-four classes.
The underlying datasets can be downloaded from 'https://cs.toronto.edu/~kriz/cifar.html'. They consist of different classes, separated into training and test images with corresponding labels. Each training and test image is also divided into several batches.