[Pycon] [new paper] "Alessandro Re" - PyTorch from the ground up

[info a pycon.it]

Title: PyTorch from the ground up
Duration: 240 (includes Q&A)
Q&A Session: 0
Language: en
Type: Training

Abstract: In this workshop we will learn how PyTorch can be used for deep learning tasks and more in general in differentiable programming contexts. We will start from the basics, introducing the package in a practical fashion, seeing how it automatically provides derivatives of a function and how this enables a gradient-based learning process. We will start with introductory, but still useful, examples, such as linear and logistic regressions and we will shift towards more advanced concepts, such as convolutional neural networks, using pre-trained models to quickly get to solve real-world problem, fine-tuning of a model and transfer learning to use the knowledge acquired in one task in new ways. We will also touch upon recurrent neural networks, widely used for natural language processing and reinforcement learning tasks.

This workshop assumes you have some basic notions of Python programming and calculus. We will use the current PyTorch 1.0 release and the suggested idioms to write correct and interoperable PyTorch code, taking advantage of the large ecosystem of tools provided by the PyTorch community. Having a GPU is not required, but can be helpful in few cases.

We will also assume that you will have a working Python 3 installed on your laptop (we wil use Python 3.7, but 3.6 will do) with the following packages installed.

 - pytorch >= 1.0 (see pytorch.org for details on installation)
 - torchvision >= 0.2.1
 - matplotlib >= 3.0
 - numpy >= 1.16.0
 - jupyter notebook >= 5.0 (or jupyterlab, if you prefer it)
 - Pillow >= 5.4
 - scikit-learn >= 0.20.2
 - tqdm >= 4.29.1

You can use Pipenv, pip or conda to install all these packages (we suggest to do so inside an environment).

We will NOT go over the installation of these tools during the workshop, because it can be very time consuming. There is plenty of information and help online on how to setup such an environment. If you are attending the workshop, we suggest you to setup your environment as soon as possible to avoid last-minute issues.

Tags: [u'machine-learning', u'deep learning', u'Pytorch', u'neural network']