[MA] Design and Inspection of a Software Framework to Explore Unsupervised Learning in Computer Vision

The motivation for Unsupervised Learning (UL) in computer vision is to learn pat- terns and structures in data without the need for labeled examples. Unsupervised learning is useful when there is no predefined target attribute to learn from1. This approach is particu- larly appropriate for problems with a large number of elements with different representations. Unsupervised learning algorithms are significant in deep learning schemes used in computer vision problems1. By using unsupervised learning, computer vision systems can learn to recog- nize patterns and features in images, which can be useful for tasks such as image classification and object detection.

W.r.t. the upcoming importance of data-oriented medicine and other applications, especially images, it would be desirable to utilize UL to find and extract knowledge from existing and not annotated data. As every use case implies different requirements and contraints, as well as UL demands for a complex data pre- and postprocessing the optimal Machine Learning tech- nology and configuration remains unknown and must be detected in a systematic manner. To include a potentially large number of UL technology, a testing framework should be designed and evaluated to analyze a given UL use case in different ways in an automated manner and enable the comparison of the different ML results.

Betreuer: Karsten Wendt