Udacity

Simple Convolutional Neural Network using TensorFlow

The goal here is to practice building convolutional neural networks to classify notMNIST characters using TensorFlow. As image size become bigger and bigger, it become unpractical to train fully-connected NN because there will be just too many parameters and thus the model will overfit very soon. And CNN solve this problem by weight sharing. We will start by building a CNN with two convolutional layers connected by a fully connected layer and then try also pooling layer and other thing to improve the model performance.

Regularization for Multi-layer Neural Networks in Tensorflow

The goal of this assignment is to explore regularization techniques.

Using TensorFlow to Train a Shallow NN with Stochastic Gradient Descent

The goal here is to progressively train deeper and more accurate models using TensorFlow. We will first load the notMNIST dataset which we have done data cleaning. For the classification problem, we will first train two logistic regression models use simple gradient descent, stochastic gradient descent (SGD) respectively for optimization to see the difference between these optimizers.

Simple Image Recognition using NotMNIST dataset

Today we're going to do some simple image recogintion using NotMNIST dataset. But before creating model for prediction, it's more important to explore, clean and normalize our dataset in order to make the learning go smoother when we actually build predictive models.