Course Information
Registration period
Year-round Recruitment
Course Level
Study Mode
Duration
6 Hour(s) 1 Minute(s)
Language
English
Taught by
Martin Musiol
Rating
3 views
Course Overview
How the next milestone in machine learning will improve the products we build
Recently, we have seen a shift in AI that wasn't very obvious. Generative Artificial Intelligence (GAI) - the part of AI that can generate all kinds of data - started to yield acceptable results, getting better and better. As GAI models get better, questions arise e.g. what will be possible with GAI models? Or, how to utilize data generation for your own projects?
In this course, we answer these and more questions as best as possible.
There are 3 angles that we take:
Application angle: we get to know many GAI application fields, where we then ideate what further projects could emerge from that. Ultimately, we point to good starting points and how to get GAI models implemented effectively.
The application list is down below.
Tech angle: we see what GAI models exist. We will focus on only relevant parts of the code and not on administrative code that won't be accurate a year from now (it's one google away). Further, there will be an excursion: from computation graphs, to neural networks, to deep neural networks, to convolutional neural networks (the basis for image and video generation).
The architecture list is down below.
Ethical angle/ Ethical AI: we discuss the concerns of GAI models and what companies and governments do to prevent further harm.
Enjoy your GAI journey!
List of discussed application fields:
Cybersecurity 2.0 (Adversarial Attack vs. Defense)
3D Object Generation
Text-to-Image Translation
Video-to-Video Translation
Superresolution
Interactive Image Generation
Face Generation
Generative Art
Data Compression with GANs
Domain-Transfer (i.e. Style-Transfer, Sketch-to-Image, Segmentation-to-Image)
Crypto, Blockchain, NFTs
Idea Generator
Automatic Video Generation and Video Prediction
Text Generation, NLP Models (incl. Coding Suggestions like Co-Pilot)
GAI Outlook
etc.
Generative AI Architectures/ Models that we cover in the course (at least conceptually):
(Vanilla) GAN
AutoEncoder
Variational AutoEncoder
Style-GAN
conditional GAN
3D-GAN
GauGAN
DC-GAN
CycleGAN
GPT-3
Progressive GAN
BiGAN
GameGAN
BigGAN
Pix2Vox
WGAN
StackGAN
etc.
Course Content
- 10 section(s)
- 39 lecture(s)
- Section 1 Introduction
- Section 2 Discriminative vs. Generative AI
- Section 3 Why does Generative AI matter?
- Section 4 Where is Generative AI located?
- Section 5 The Power of Generative Adversarial Networks (GANs)
- Section 6 Why did it take them so long?
- Section 7 The implementation of a simple GAN
- Section 8 A Deep-Dive into Various Application Fields
- Section 9 Concerns around Generative AI Models
- Section 10 Noteworthy GAN Architectures
What You’ll Learn
- How to implement Generative AI models. We focus on proper concept implementation and relevant code (no administrative code)
- Get to know the broad spectrum of GAI applications and possibilities tangibly eg. 3D object generation, interactive image generation, and text generation
- How to identify great ideas in the GAI space and make best use of already developed models for realising your projects and ideas
- How to augment your dataset such that it ultimately improves your machine learning performance eg. for classifiers of rare diseases
- Learn about the ethical side: what are the concerns around GAI, incl. deep fakes, etc.
- The technical side: from the evolution of generative models, to the generator-discriminator interplay, to common implemenation issues and their remedies
Skills covered in this course
Reviews
-
SSriram Vadlamani
Good high-level introductionl.
-
SSnehal sadashiv bodke
It was good ,very informative,introduced new terms related to AI.
-
RRahul Saha
Sharable PDF document with the important links and slide content would have been great.
-
SSantiago Arroyo Marioli
Too much introduction to get to the interesting part. Even though I would have preferred to go more in depth of the topics the notebooks are there for you to work with. Too many topics to summarize it in 6hs, maybe is better to have less topics and go deeper. For example, there aren't any suggestions on how to actually build the models, what things could go wrong, what is crucial, what isn't, etc.
