This thesis explores new applications of Unsupervised Curriculum Learning (UCL) across domains beyond the previously explored area of remote sensing imagery. UCL is a framework for training deep learning image classification models without needing labeled data. The framework consists of several components, including a convolutional neural network, a clustering algorithm and a selection operation. The machine learning techniques in these components interact to find meaningful patterns in the data. Some of the techniques and the interaction between them are explored further to improve the classification performance of the trained models. The updates to UCL investigated in this thesis include spectral clustering, outlier detection, and dimensionality reduction. These techniques increased the classification performance from 68% to 94% f1-score on a three-class subset of the CIFAR-10 dataset and from 68% to 75% on a five-class subset. When evaluating the updated UCL for medical diagnosis, scene recognition and historical document classification, the trained models reached f1-scores of 85%, 82% and 62% respectively, highlighting both strengths and weaknesses of the methodology. The results show the viability of using UCL on a broader range of applications but also reveal some potential areas of improvement.