Deep Learning Tutorial with PyTorch

Deep learning tutorial with PyTorch has been initiated to teach undergraduate students in SeoulTech about basic concepts of deep learning and hands-on practices with PyTorch. Even though there are so many good lectures and tutorials on deep learning, I would like to share my humble slides and examples because I also have learned many things from open and public lectures, tutorials, and articles. I hope that my slides and examples are also helpful to others.

• Download my tutorial slides
• Download my practice examples in a ZIP file

This tutorial is the last part of my four sequels of tutorials. Many examples and practices are connected each other. Please refer its neighborhood tutorials also.

1. Python Review: From Beginners to Intermediate
2. Programming meets Mathematics: Building Mathematical Intuition with SciPy
3. Machine Learning Tutorial with Scikit-learn
4. Deep Learning Tutorial with PyTorch

Practice Examples

Source codes are enumerated in the order of my tutorial slides.

• PyTorch
  • Creating a Tensor
  • Reshaping a Tensor
  • Line Fitting from Two Points
  • CPU vs. GPU-acceleration
  • Automatic Differentiation
  • Automatic Differentiation - More Analysis
  • Gradient Descent by Hands
  • Gradient Descent by torch.optim
• Neural Network
  • Visualizing Activation Functions
  • Observing Vanishing Gradient - Sigmoid
  • Observing Vanishing Gradient - ReLU
  • The Iris Flower Dataset [UCI ML Repository]
    • Iris Flower Classification - No Class
    • Iris Flower Classification - My Style
• Convolutional Neural Network
  • The MNIST Dataset [homepage]
    • Loading the MNIST Dataset
    • Digit Classificaiton with the MNIST Dataset
    • Loading My Network and Testing It
    • Different Styles for NN Classes
  • Visualizing Learning Rate Schedulers
• Recurrent Neural Network
  • The Name2Lang Dataset [homepage]
    • Name2Lang Classification with a Character-level RNN

NOTE: All examples contain their basic NN architectures and hyperparameters. One of main objectives in practices will be their performance improvement by changing the architectures and selecting hyperparameters.

Author

• Sunglok Choi (sunglok AT seoultech DOT ac DOT kr)