This repository contains Jupyter notebooks used for tutoring students in the Neural Networks and Deep Learning courses during the academic year 2023/2024. The tutorials cover various topics related to the usage of PyTorch, a popular deep learning framework.
The repository includes the following notebooks:
Lesson 0 - Introduction to Python: An overview of Python for those who have never coded before
Lesson 1 - Pytorch: This notebook covers the main features of Torch, including its basic operations, tensor manipulation and autograd
Lesson 2 - Building Neural Networks: We cover PyTorch's torch.nn module, with topics such as defining network architectures, implementing forward and backward passes, and optimizing models with different optimization algorithms
Lesson 3 - Convolutional Neural Networks: We dive into the world of CNNs with PyTorch. This notebook demonstrates how to create CNN architectures for image classification tasks. In it, we also talk about the difference between using GPUs vs CPUs and how to perform data augmentation.
Lesson 4 - Setting up a project: In this notebok we explain how to find a dataset and upload it in PyTorch, how to tune your model hyperparameters and how to create a saliency map
Lesson 5 (bonus) - Autoencoders: A shorter lesson with the basics of how to implement and train a simple autoencoder for MNIST digits
You can either run these notebooks on Google Colab or download them to your local machine and execute them using Jupyter Notebook or JupyterLab. To run them locally, ensure you have Python installed along with the necessary dependencies (PyTorch, NumPy, Matplotlib, etc.).
We welcome any feedback or contributions to improve these tutorials. Feel free to open an issue or submit a pull request with your suggestions, corrections, or additional content.
The tutorials were created by Federico Milanesio and Davide Pirovano at UniTO, both PhD students currently working in neural networks and deep learning.
Federico Milanesio - federico.milanesio@unito.it Davide Pirovano - davide.pirovano@unito.it
© 2024. This work is openly licensed via CC BY-NC-SA 4.0 DEED