COMP551_Project4

Adding Gaussian noise to gradient values of back propagation in order to make "Differential Privacy"

---

The code for implementing MLP is taken from Theano's tutorial page

---

• Use "mlp1.py" to train the original MLP model for MNIST dataset.
• Put "mnist.pkl.gz" file in your code directory
• No noise added to this model

---

• Use "mlp1_dp.py" to adding different level of noise to model
• First we clip gradinet values
• We add different level of Gaussian to Gradient values
• keep a copy of "mnist.pkl.gz" file in your code directory

---

Use "plotter.py" to plot your final results