Introduction to Generative Adversarial NetworksLaajuus (3 op)

Osaamistavoitteet

Knowledge and understanding:
The students will learn some basic concepts and algorithms of Generative Adversarial Networks (GANs) for image generation, and they will understand how to implement GANs in Python. Students will gain knowledge of how neural networks (NN) and convolutional neural networks (CNN) are used in GANs. Some applications of GANs will be introduced.

Skills:
The students will be able to use Google Colaboratory through the experiments with GAN. They will also be able to build a GAN model with Python.

Practice and Lab work:
Based on sample programs, students will learn how to develop neural network models including a convolutional neural network and a transposed convolutional neural network (tCNN) with Python. Students will finally develop a GAN model to generate realistic handwritten images and demonstrate how to generate fake face images.

Sisältö

1. Introduction to GAN and its structure and its applications. (Lecture)
2. Understanding the function of a discriminator in GAN by studying binary classification with NN and CNN models. (Lecture + Lab work)
3. Studying the function of a generator in GAN by using a NN model and tCNN with a generation of handwritten digits from white noise images by GAN. (Lecture + Lab work)
4. Building a GAN model by combining discriminator and generator networks. (Lecture + Lab work)
5. Demonstration of deep convolutional generative adversarial networks (DCGANs). (Lecture + Lab work)
6. Generating full color human face images by GAN. (Lecture + Lab work)

Esitietovaatimukset

It is preferable to have a basic programming skill in Python and to be able to use basic data structures and algorithms.

Arviointikriteerit, tyydyttävä (1)

Students know concepts and a structure of GANs.
Students can understand programming about GANs.
Students can know the applications of GAN.

Arviointikriteerit, hyvä (3)

Students can understand concepts and a structure of GANs.
Students can understand a GAN model by programming.
Students can know and explain the applications of GAN.

Arviointikriteerit, kiitettävä (5)

Students can understand and explain concepts and the structure of GANs.
Students can understand and build a GAN model by programming.
Students can know and explain the applications of GAN and have own idea about an original application.

Arviointikriteeri, hyväksytty/hylätty

Students know concepts and a structure of GANs.
Students can understand programming about GANs.
Students can know the applications of GAN.