Feature-Extraction

Can extracting features on the last layer of a CNN help to battle a lack of training data? This is a rough experiment to try to answer that question. The CIPHAR10 data set is used, with both a 'simple' CNN and a slightly more complex CNN.