e7dud7e / deeplearningnd

Deep Learning nanodegree - Udacity

Deep Learning Foundations Nanodegree

Overview of projects

This is an overview of deep learning projects that I completed for the Udacity Deep Learning Nanodegree.

Install

These projects use Python 2 or 3 with TensorFlow 1.0 or higher. See the individual README in each project folder for requirements.

Feed Forward Neural Networks built in Numpy

• Built fully connected layers and implemented back propogation in numpy. Used to predict bike share usage.

Convolutional Neural Network for image classification

• Built a convolutional neural network and optimized with batch normalizaiton to reach 81.8% test accuracy.

Recurent Neural Network for TV script generation

• Built a recurrent neural network to generate text that resembles tv scripts from the Simpsons TV show.

RNN for Sentiment Analysis

• Built a recurrent neural network of multiple LSTM (long-short-term memory) cells.
• Used an embedding layer to convert words to vectors.
• Used GloVe (global vector of words) to set pre-trained word vectors in embedding layer.

Seq2Seq RNN for language translation

• Built a seq2seq recurrent neural network to translate text from English to French.

Generative Adversarial Network for new image generation

• Built a generative adversarial network to generate new images that resemble human faces. The discriminator learns to distinguish between real photos of faces and non-faces. Concurrently, the generator network learns to generate data that the discriminator considers to be real faces.