Introduction to deep learning with PyTorch

10:00 – 12:00

14:30 – 19:30

10:00 – 11:45

by Martin Tamajka from Faculty of Informatics and Information Technologies of Slovak Technical University in Bratislava

How does the computer distinguish between cats and dogs in the picture? In the last few years, deep learning has become the source of both excitement and great expectations. In this lab you will get both theoretical and practical introduction to deep learning – covering the very basics of neural networks, multi-layer networks and in the end classification of images using convolutional neural networks. The goal we follow at the workshop is simple – it is easy to just start coding neural networks, thanks to frameworks like PyTorch or Tensorflow, but with understanding of the inner mechanisms the life is much easier.

In this course we will cover two very important tasks of computer vision – classification and segmentation of images. In the first part, we’ll provide a brief recapitulation of machine learning. Next, we’ll cover the very basics of neural networks, starting from a single neuron and advancing to multilayer networks. In the end of this part we’ll train a multilayer perceptron to classify 2D images.

The next part will be dedicated to convolutional neural networks. We’ll cover basic concepts and building blocks of this type of neural networks. In the end of this part we’ll improve the performance of classification of 2D images by using convolutional networks and advance to classification of 3D volumetric medical data.

The last part of the course will cover a special type of neural network – autoencoders. We’ll introduce some common use cases of this type of networks. As hands-on exercise we’ll use a special kind of autoencoder to perform a fully automatic image segmentation.

There’s one more thing – a competition. Course participants will have the opportunity to participate in a mini-competition, in which their task will be to achieve the highest possible accuracy within a specific classification or segmentation task by the next day. The competition will be evaluated on Sunday morning.

The workshop will be held virtually and for free at our Discord server. You can join at https://discord.gg/hMSA6Gr. The link will be valid until Friday 01 May, midnight. Please use your real name, unknown users will not be accepted.