Workshop: Generative AI with Diffusion Models

Friday 18 Oct 2024, 10:00 → 18:00 Europe/Ljubljana

MS Teams

Domen Verber, Jani Dugonik

Description: At the workshop, we will present one of the most advanced areas of artificial intelligence - generative artificial intelligence. We will get to know all the components of generative diffusion models in more detail and in several successive steps we will create a test application for generating images based on text description.

The workshop is intended for both academic researchers and practitioners in the field of artificial intelligence who want to delve deeper into generative models and use them in their work. Throughout the day, we will explore concepts, conduct practical exercises and discuss the latest developments in this field.

After completing the workshop, participants will be able to obtain official certification from the NVIDIA Deep Learning Institute.

Detailed description: Thanks to the increasing capabilities of modern computer systems and advances in scientific theory, generative AI models have become more accessible than ever. Generative AI will play an important role in all branches of industry due to many applications such as creative creation of new content, rewarding existing data, simulations and planning, anomaly detection, new drug discovery, personalized recommendations and much more. In this workshop, we will delve into diffusion models for converting text to images.

The workshop will take place in the AWS cloud environment prepared by NVIDIA. Interactive access to powerful computing resources via a browser will be enabled. Each participant will have a powerful graphics accelerator at their disposal. The practical part of the workshop will be carried out using interactive learning documents, using the Python programming language and the PyTorch software library. The acquired knowledge can be easily transferred to other development environments.

After completing the workshop, participants will be able to take a short test and an additional programming task to earn official certification from the NVIDIA Deep Learning Institute. It will be possible to carry out the test and obtain the certificate at least six months after the end of the workshop.

Workflow: The workshop will take place in an interactive cloud environment with access via a browser.

Difficulty: Advanced

Language: English

Suggested prerequisites: Advanced knowledge of the Python programming language and programming libraries for building machine learning models. More detailed knowledge of the theoretical and practical aspects of deep neural networks.

Target audience: Students, academics and practitioners interested in generative models of artificial intelligence

Skills to be gained:

• Hands-on experience in image generation from noise using U-Net neural network
• Understanding the generative diffusion process based on denoising to improve the quality of the generated image
• Understanding the differences between the diffusion process based on denoising and other generative models
• Knowledge of procedures for controlling generated content by including context in the diffusion generative process

Max. number of participants: 30

Virtual location: MS Teams

Organisers:

---

Lecturers:

Name:	Domen Verber
	Domen Verber is an assistant professor at the Faculty of Electrical Engineering and Computer Science of the University of Maribor (UM FERI) and ambassador of the NVIDIA Deep Learning Institute for the University of Maribor and their HPC specialist. He has been dealing with HPC and artificial intelligence issues for more than 25 years.
	domen.verber@um.si, deep.learning@um.si

 

Name:	Jani Dugonik
	Jani Dugonik is an academic researcher at the Faculty of Electrical Engineering, Computer Science and Informatics of the University of Maribor (UM FERI). He has been working in the field of natural language processing and evolutionary algorithms for more than 10 years.
	jani.dugonik@um.si

---

10:00 → 10:15 Introduction: Meet the instructor; Create an account at courses.nvidia.com/join

10:15 → 11:15 From U-Nets to Diffusion: Build a U-Net, a type of autoencoder for images; Learn about transposed convolution to increase the size of an image; Learn about non-sequential neural networks and residual connections; Experiment with feeding noise through the U-Net to generate new images

11:15 → 11:25 Break

11:25 → 12:25 Control with Context: Learn how to alter the output of the diffusion process by adding context embeddings; Add additional model optimizations such as Sinusoidal Position Embeddings, The GELU activation function, Attention

12:25 → 13:25 Text-to-Image with CLIP: Walk through the CLIP architecture to learn how it associates image embeddings with text embeddings; Use CLIP to train a text-to-image diffusion model

13:25 → 14:25 Break

14:25 → 15:25 State-of-the-art Models: Review various state-of-the-art generative ai models and connect them to the concepts learned in class; Discuss prompt engineering and how to better influence the output of generative AI models; Learn about content authenticity and how to build trustworthy models

15:25 → 16:25 Final Review: Review key learnings and answer questions; Complete the assessment and earn a certificate; Complete the workshop survey; Learn how to set up your own AI application development environment.