Deep Learning assignments

Homeworks from the Deep learning course, implementations done with Tensorflow
Jupyter Notebook files contain code with explanations.

Hw1:

Underfitting and overfitting problem for the polynomial regression

Hw2:

Back-propagation for 100-layer network, investigation of conditions under which the training fails

Project1:

1. Classification on the toy dataset with two classes
2. Transfer learning from MNIST classification of 9 digits to 10 digits
3. PCA on 100 test samples of MNIST classifier

Project2:

1. Q learning in the linear environment
2. Q learning for the 4-legged spider
3. Deep Q Network for the Breakout environment

Project3:

1. Training a simple neural network playing tic-tac-toe. Proved that the agent can solve any game with all behaviours (while wins agains the random player are not the sufficient conditions, since it gives only the probabilistic conclusion).
2. Mini AlphaGo Zero: train 5 neural networks such that the first is trained based on purely random plays, the second is trained based on game records played by the first neural network, the third is trained based on game records played by the second neural network and so on.