Generative AI Theory

(1) Course Outline

The objective of the course is to deepen understanding of how the models covered in the Introduction to Generative AI are actually learned, and papers on important deep learning methods that students interested in deep learning, not limited to generative AI, should learn. Each session will focus on one paper at a time, with in-depth explanations and, in some cases, demonstrations and applications. The topics are image generation, natural language processing, and multimodal language processing. Students who have already taken “Introduction to Generative AI” or have equivalent knowledge are assumed to attend this course.

(2) Contents of each session

Session 1: Image Generation 12/13

The course introduces the process of learning classical GAN and the newest method, Stable Diffusoin, in terms of image generation, based on mathematical formulas.

 Session 2: Natural Language Processing and Multimodal Language Processing 12/20

Transformer [Vaswani+ NeurIPS17], a breakthrough in the field of natural language processing, will be introduced. Specifically, we will explain Attention, the model structure and formulas, and introduce (+demo) Transformer applications up to the present, including BERT.