If you are viewing this file in preview mode, some links won't work. Find the fully featured Jupyter Notebook file on the website of Prof. Jens Flemming at Zwickau University of Applied Sciences. This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Project: QMNIST with CNNs

Solve the QMNIST digit recognition task with a CNN and analyze the chosen CNN.

Work package 1: Training

Train a suitable CNN. Do not (!) use PCA or other feature reduction techniques. Do not (!) crop the images because convolutions need some space around the interesting parts of an image.

Try to get at least 85 percent correct classifications on the test set (without using the test set for training, of course).

Save your model.

Work package 2: Analysis

Visualize filters for the first convolutional layer in your CNN and try to interpret some of them (what features do they detect).

Take an image correctly classified to show a zero and modify it slightly to make you CNN 'think' that it's a five although human eye clearly sees a zero. You may use code from the lectures.

Solution

• training.ipynb (2 points, one for code, one for 85 or more percent correct classifications)
• analysis.ipynb (2 points, one for filter visualization, one for modified zero image)
• saved model (directory created when saving the model)

Zip the two files and the directory, name the zip file forename_surname.zip, and send it to the lecturer.