emilio-molina / deep_learning_nanodegree

My solutions for the Deep Learning Nanodegree (https://eu.udacity.com/course/deep-learning-nanodegree-foundation--nd101)

Deep Learning Nanodegree Foundations

Author: Emilio Molina ([e-mail])

This repository contains:

• My solutions for the Deep Learning Nanodegree Foundations: https://eu.udacity.com/course/deep-learning-nanodegree-foundation--nd101
• A brief summary of all projects (in this README.md)

What I've learned in the nanodegree

A general overview on deep learning, and some hands-on experience with exercises and projects using Tensorflow. I think the content of the nanodegree can be summarized with the 5 delivered projects (and manually reviewed by Udacity).

Projects

The solutions can be found in the different folders inside this repository. A brief summary:

1. Your first neural network

In this project you implement your own neural network from scratch using numpy. You implement not only the forward pass (prediction) but also the backpropagation algorithm (training). It is a very simple use case with a single hidden layer, but you obtain very interesting results (see figures).

2. Image classification using convolutional neural networks

Very interesting project where you use convolutional neural networks for implementing a image classifier (using CIFAR-10 dataset). Example of the achieved results:

3. Generate TV scripts using RNN

You use a recurrent neural network trained on TV scripts of The Simpsons to automatically generate a new one. This is the kind of script automatically generated:

4. English-to-French translation using deep learning

You use a recurrent neural network to implement a sequence-to-sequence system for language translation. Concretely, you use a dataset of english-to-french translation. Example of translation performed by the system:

5. Digits and faces generation using generative adversarial networks

This project is quite nice to understand GANs, and at the end of it you have a quite nice convolutional GAN able to generate almost-realistic faces. Check the results: