Deep Learning Foundations

Notes from the 2017 course

Bike Sharing Data

Image Classification

Basic CNN with flatten and dense network head to output softmax probabilities of image classes.

TV Script Generator

Machine Translation

RNN with LSTM to encode word2vec input sequence and decode resulting hidden state.

Generating Human Faces

Develop a GAN to learn to draw faces from a collection of celebrity headshots.

Course Opinions

There is a lot of hand holding. Jupyter notebooks are very well organized, but teaching TensorFlow is more the idea than overall deep learning content.

Suggested improvements include more quizzes and exams or a capstone project from Kaggle. Also, should offer some software engineering best practices which Jupyter notebooks typically avoid altogether.