Deep Learning with PyTorch

aka

How not to waste your time in coding artificial neural networks

This repository contains the resources for some practical lessons about the Deep Learning part of the Statistical Machine Learning course, held by Prof. Luca Bortolussi for the MSc students in Data Science and Scientific Computing at the University of Trieste.

My email: francesco.cicala00@gmail.com

Lectures index (approximately):

• Lesson 1:

• • Why do we need a deep learning library
  • PyTorch: an overview
  • Static graphs, dynamic graph, automatic differentiation
  • Tensors
  • Autograd

• Lesson 2:

• • Linear model for spiral data
  • NN for spiral data
  • Importing CIFAR10
  • Classification on CIFAR10 with a FCNN
  • torch.nn, nn.Module, nn.functionals, optim

• Lesson 3:

• • Convolutional layers
  • Pooling layers
  • Simple CNN (scrambled vs not scrambled)
  • Derive new blocks from nn.Module
  • Saving and loading a model

• Lesson 4:

• • A simple recurrent cell from scratch
  • Create your Dataset
  • Learning rate scheduler
  • Image classification with a RNN

• Lesson 5:

• • Uploading your own dataset with ImageFolder
  • Data augmentation through on the fly random transformations
  • Importing a pretrained model
  • Early stopping
  • Fine tuning