In this exercies you will try to build some very specific neural networks "by hand" and try some normalization. I hope it will build some intuition on how neural networks and "Non-linearities" work.

The exercies is solved by writing code in hand_trained_neural_networks.py and distorted_cifar. In hand_trained_neural_networks your are not supposed to alter the function neural_net_from_parameters.

In hand_trained_neural_networks.py you may need to change the number of weight-matrices sendt to the function neural_net_from_parameters.

Download the code with:

git clone https://github.com/sigmunjr/INF5860_uke5.git

To check if a solution is correct you can run (may take some time)

nosetests

To check a tests in a single file run e.g.:

nosetests test_hand_trained_neural_networks.py

To run a single_test:

nosetests test_hand_trained_neural_networks.py:test_grayscale_shift

Solutions can be found on github:

git clone https://github.com/sigmunjr/INF5860_uke5_solutions.git

Good luck!