Use TensorFlow to access selected Udacity assignments. Open the corresponding iPython notebook and follow the instructions to implement and run each indicated step.

Prepare a webpage reporting the results.

  1. Assignment 1: notMNIST (due by: 22/08)
    Preprocess notMNIST data and train a simple logistic regression model on it
    You should install the Python Imaging Library (PIL) required to load image files.

  2. Assignment 2: Stochastic Gradient Descent (due by: 29/08)
    Train a fully-connected network using Gradient Descent and Stochastic Gradient Descent
    Your new model should perform better than the one you developed for Assignment 1. Also, the time required to train using Stochastic Gradient Descent (SGD) should be considerably less than simple Gradient Descent (GD).

  3. Assignment 3: Convolutional Models (due by: 24/10)
    Design and train a Convolutional Neural Network
    Improve the model by experimenting with its structure - how many layers, how they are connected, stride, pooling, etc. For more efficient training, try applying techniques such as dropout and learning rate decay. What does your final architecture look like?

Copyright © Luiz Velho