udacity/deep-learning-v2-pytorch

★ 5,504Jupyter NotebookAudience · researcherComplexity · 3/5Setup · moderate

Mindmap

mindmap
  root((repo))
    What it does
      Course notebooks
      Deep learning basics
      Project starters
    Topics
      Neural networks
      Image classifiers
      Text generation
      Face generation
    Tech Stack
      Python
      PyTorch
      Jupyter Notebook
      Anaconda
    Projects
      Sentiment classifier
      Style transfer
      GAN faces
    Audience
      Students
      Beginners

mindmap root((repo)) What it does Course notebooks Deep learning basics Project starters Topics Neural networks Image classifiers Text generation Face generation Tech Stack Python PyTorch Jupyter Notebook Anaconda Projects Sentiment classifier Style transfer GAN faces Audience Students Beginners

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Work through structured exercises to learn how to build and train image classifiers with PyTorch, starting from handwritten digits

USE CASE 2

Use the project starter notebooks to build a sentiment classifier that reads movie reviews and labels them positive or negative

USE CASE 3

Train a generative adversarial network to generate synthetic human faces using the provided celebrity photo dataset project

Tech stack

PythonPyTorchJupyter NotebookAnaconda

Getting it running

Difficulty · moderate Time to first run · 1h+

Requires Anaconda installation and environment setup, GPU access speeds up training projects but is not strictly required for all exercises.

In plain English

This repository holds the tutorial notebooks and project starter code for Udacity's Deep Learning Nanodegree program. Deep learning is a branch of machine learning where software learns to recognize patterns by passing data through many layers of mathematical operations, loosely inspired by how neurons connect in a brain. The course uses PyTorch, a popular toolkit that makes it practical to build and train these kinds of models in Python. The material is organized into several topic areas. The neural networks section starts with the basics: how a network learns by adjusting its internal numbers in response to mistakes, illustrated through a project that predicts student admissions. From there, the course covers convolutional neural networks, which are particularly good at recognizing things in images. Exercises here include classifying handwritten digits and photos of common objects, as well as a style transfer project where a network applies the visual style of one image (like a painting) onto a different photo. The recurrent networks section covers models that process sequences, such as chains of words or time-series data. One exercise trains a network to generate new TV show scripts in the style of Seinfeld. Another builds a sentiment classifier that reads a movie review and predicts whether it is positive or negative. There is also material on word embeddings, which are a way of representing words as numbers so that words with similar meanings end up near each other mathematically. The generative adversarial networks section teaches how to train two networks against each other so that one learns to generate realistic-looking images. Projects here include generating synthetic human faces using a celebrity photo dataset. Projects in the program are reviewed by human graders at Udacity, and the starting code for each project is included in this repository. Setup requires installing Anaconda, a Python environment manager, and the README includes a step-by-step guide for getting the dependencies installed on Linux, Mac, or Windows.

Copy-paste prompts

Prompt 1

I am working through the Udacity deep learning course and want to understand how backpropagation updates weights. Show me how gradient descent works step by step using the student admissions notebook as context.

Prompt 2

How do I install the Anaconda environment and all dependencies needed to run the Udacity deep-learning-v2-pytorch notebooks on a Mac?

Prompt 3

I want to complete the style transfer project. Explain how the VGG network extracts content features and style features from two images and how the loss is computed.

Prompt 4

Show me how to train an LSTM network to generate new Seinfeld TV show scripts using the recurrent networks notebook from this course repo.

Open on GitHub → Explain another repo

← udacity on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.