Deep Learning Foundation

Welcome to the Deep Learning Nanodegree Foundations Program! In this lesson, you'll meet your instructors, find out about the field of Deep Learning, and learn how to make the most of the resources Udacity provides.

Anaconda is a package and environment manager built specifically for data. Learn how to use Anaconda to improve your data analysis workflow.

Learn how to use Jupyter Notebooks to create documents combining code, text, images, and more.

In this lesson, you'll get your hands dirty by playing around with a few examples of deep learning. Don't worry if you don't understand what's going on! The goal here is just for you to play around with some models others have already created and have fun.

Learn about linear regression and logistic regression models. These simple machine learning models are the building blocks of neural networks.

In this lesson, you'll review the matrix math you'll need to understand to build your neural networks. You'll also explore NumPy, the library you'll use to efficiently deal with matrices in Python.

In this lesson, you'll dive deeper into the intuition behind Logistic Regression and Neural Networks. You'll also implement gradient descent and backpropagation in python right here in the classroom.

In this project, you'll build and train your own Neural Network from scratch to predict the number of bikeshare users on a given day. Good luck!

Project Description - Your first neural network

Project Rubric - Your first neural network

In this lesson, you'll learn some of the basics of training models. You'll learn the power of testing and cross validation, and some interesting metrics to evaluate models, such as accuracy or R2 score.

In this lesson, Andrew Trask, the author of Grokking Deep Learning, will walk you through using neural networks for sentiment analysis. In particular, you'll build a network that classifies movie reviews as positive or negative just based on their text!

In this lesson, you'll get to use TFLearn, an incredibly simple library for building neural networks. You'll get to apply TFLearn on both a sentiment analysis assignment as well as an image classification assignment.

In this lesson, we'll cover representing words as inputs to a network. We'll cover Bag of Words as well as Word2Vec. In addition, we'll share a few resources to introduce RNNs and LSTMs.

In this lesson, you'll build your own small version of TensorFlow, called MiniFlow. By building this, you'll gain an understanding of how TensorFlow works under the hood, and gain more insights into important concepts like backpropagation.

Take advantage of Amazon's GPUs to train your neural network faster. In this lesson, you'll setup an instance on AWS and train a neural network on a GPU.

Vincent Vanhoucke, Principal Scientist at Google Brain, introduces you to deep learning and Tensorflow, Google's deep learning framework.

Vincent walks you through how to go from a simple neural network to a deep neural network. You'll learn about why additional layers can help and how to prevent overfitting.

Vincent explains the theory behind Convolutional Neural Networks and how they help us dramatically improve performance in image classification.

Siraj will go over the history of image classification, then he'll dive into the concepts behind convolutional networks and why they are so amazing.
He'll also create an Image Classifier for cats & dogs in 40 lines of Python!

Classify images from the CIFAR-10 dataset using a convolutional neural network.

Project Description - Image Classification

Project Rubric - Image Classification

Recurrent neural networks are able to learn from sequences of data. In this lesson, you'll learn the concepts behind recurrent networks and see how a character-wise recurrent network is implemented in TensorFlow.

Siraj is going to predict the closing price of the S&P 500 using a special type of recurrent neural network called an LSTM network. He'll explain why we use recurrent nets for time series data, and why LSTMs boost our network's memory power.

In this lesson, we'll look at number of different hyperparameter that are important for our deep learning work. We'll discuss starting values and intuitions for tuning each hyperparameter.

In this lesson, you'll learn about embeddings in neural networks by implementing the word2vec model.

Siraj will show you how to generate art with deep learning networks. He'll also talk about the history of computer generated art, and why deep learning models are so good at making art.

A Q&A Session with Sai Soundararaj and Naren Thiagarajan from FloydHub.

In this lesson you'll learn about using TensorBoard to inspect your network. You can view the TensorFlow graph and the distributions of variables in the network. You can also use it to compare multiple training runs with different hyperparameters.

Siraj is going to build a music generating neural network trained on jazz songs in Keras. He'll go over the history of algorithmic generation, then he'll walk step by step through the process of how LSTM networks help us generate music.

Siraj will teach you how to transform an essay into a single sentence using Keras, and talk about word embeddings, encoder-decoder architectures and the role of attention in the theory of learning.

In this lesson, you'll learn how to find good initial weights for a neural network. Having good initial weights can place the neural network close to the optimal solution. This allows the neural network to come to the best solution quicker.

Implement a sentiment prediction RNN

Generate a TV script using a recurrent neural network.

Project Description - Generate TV Scripts

Project Rubric - Generate TV Scripts

In practice, most people don't train their own networks on massive datasets. In this lesson, you'll learn how to use a pretrained network on a new problem with transfer learning.

Siraj will build his own language translator using Tensorflow! He'll go over several translation methods and talk about how Google Translate is able to achieve state of the art performance.

Here you'll learn about a specific architecture of RNNs for generating one sequence from another sequence. These RNNs are useful for chatbots, machine translation, and more!

Siraj will make a Question Answering chatbot using Dynamic Memory Network. He'll go over different chatbot methodologies, then dive into how memory networks work, with accompanying code in Keras.

Here you'll learn about using reinforcement learning to train an AI agent to play games.

Siraj will show how to solve the multi-armed bandit problem (maximizing success for a given slot machine) using a reinforcement learning technique called policy gradients.

Use a neural network to translate from one language to another.

Project Description - Translation Project

Project Rubric - Translation Project

Siraj is going to build a variational autoencoder capable of generating novel images after being trained on a collection of images, using handwritten digit images as training data.

Autoencoders are neural networks used for data compression, image denoising, and dimensionality reduction. Here, you'll build autoencoders using TensorFlow.

Ian Goodfellow, the inventor of GANs, introduces you to these exciting models. You'll also implement your own GAN on the MNIST dataset.

Siraj is going to use a Deep Convolutional GAN to generate images of the alien language from the movie arrival that we can then stitch together to animate into video.

See how memory augmented neural networks can help achieve one-shot classification for a small labeled dataset. He'll also review the architecture of it's inspiration (the neural turing machine)

In this lesson, we'll look at number of different hyperparameter that are important for our deep learning work. We'll discuss starting values and intuitions for tuning each hyperparameter.

In this lesson you'll implement a Deep Convolution GAN to generate complex color images of house numbers.

Compete two neural networks against each other to generate realistic faces.

Project Description - Generate Faces

Project Rubric - Generate Faces

Ian Goodfellow leads you through a semi-supervised GAN model, a classifier that can learn from mostly unlabeled data.

Utilize your guaranteed admission and enroll into a Career-Ready Nanodegree program

Welcome to the Deep Learning Nanodegree program!

In this lesson, you'll get your hands dirty by playing around with a few examples of deep learning. Don't worry if you don't understand what's going on! The goal here is just for you to play around with some models others have already created and have fun.

Anaconda is a package and environment manager built specifically for data. Learn how to use Anaconda to improve your data analysis workflow.

Learn how to use Jupyter Notebooks to create documents combining code, text, images, and more.

In this lesson, you'll review the matrix math you'll need to understand to build your neural networks. You'll also explore NumPy, the library you'll use to efficiently deal with matrices in Python.

In this lesson, Luis will give you solid foundations on deep learning and neural networks. You'll also implement gradient descent and backpropagation in python right here in the classroom.

Mat will introduce you to a different error function and guide you through implementing gradient descent using numpy matrix multiplication.

Now that you know what neural networks are, in this lesson you will learn several techniques to improve their training.

In this lesson, Andrew Trask, the author of Grokking Deep Learning, will walk you through using neural networks for sentiment analysis.

In this section you'll get a hands-on introduction to Keras. You'll learn to apply it to analyze movie reviews.

In this section you'll get a hands-on introduction to TensorFlow, Google's deep learning framework, and you'll be able to apply it on an image dataset.

Take advantage of Amazon's GPUs to train your neural network faster. In this lesson, you'll setup an instance on AWS and train a neural network on a GPU.

In this lesson, you'll learn how to find good initial weights for a neural network. Having good initial weights can place the neural network close to the optimal solution. This allows the neural network to come to the best solution quicker.

Alexis explains the theory behind Convolutional Neural Networks and how they help us dramatically improve performance in image classification.

Autoencoders are neural networks used for data compression, image denoising, and dimensionality reduction. Here, you'll build autoencoders using TensorFlow.

In practice, most people don't train their own networks on massive datasets. In this lesson, you'll learn how to use a pretrained network on a new problem with transfer learning.

In this lesson, Sebastian Thrun teaches us about his groundbreaking work detecting skin cancer with convolutional neural networks.

In this project, you will learn how to build a pipeline to process real-world, user-supplied images. Given an image of a dog, your algorithm will identify an estimate of the canine’s breed.

Ortal will introduce Recurrent Neural Networks (RNNs), which are machine learning models that are able to recognize and act on sequences of inputs.

Luis explains Long Short-Term Memory Networks (LSTM), and similar architectures which have the benefits of preserving long term memory.

In this lesson, we'll look at number of different hyperparameter that are important for our deep learning work. We'll discuss starting values and intuitions for tuning each hyperparameter.

In this lesson, you'll learn about embeddings in neural networks by implementing the word2vec model.

Implement a sentiment prediction RNN

Ian Goodfellow, the inventor of GANs, introduces you to these exciting models. You'll also implement your own GAN on the MNIST dataset.

In this lesson you'll implement a Deep Convolution GAN to generate complex color images of house numbers.

Ian Goodfellow leads you through a semi-supervised GAN model, a classifier that can learn from mostly unlabeled data.

Reinforcement learning is a type of machine learning where the machine or software agent learns how to maximize its performance at a task.

Learn how to mathematically formulate tasks as Markov Decision Processes.

In reinforcement learning, agents learn to prioritize different decisions based on the rewards and punishments associated with different outcomes.

The dynamic programming setting is a useful first step towards tackling the reinforcement learning problem.

Write your own implementation of Monte Carlo control to teach an agent to play Blackjack!

Learn about how to apply temporal-difference methods such as Sarsa, Q-Learning, and Expected Sarsa to solve both episodic and continuous tasks.

With reinforcement learning now in your toolbox, you're ready to explore a mini project using OpenAI Gym!

Review the fundamental concepts of reinforcement learning, and learn how to adapt traditional algorithms to work with continuous spaces.

Extend value-based reinforcement learning methods to complex problems using deep neural networks.

Policy-based methods try to directly optimize for the optimal policy. Learn how they work, and why they are important, especially for domains with continuous action spaces.

Learn how to combine value-based and policy-based methods, bringing together the best of both worlds, to solve challenging reinforcement learning problems.

Build a quadcopter flying agent that learns to take off, hover and land using reinforcement learning.