CSSE/MA 490 Generative AI

Prerequisites: CSSE/MA 416 Deep Learning

Course Description:

Building upon the foundations of Deep Learning, this course is about the exciting world of Generative AI, the next frontier in artificial intelligence. While CSSE/MA 416 focused on Discriminative AI – classifying existing data (e.g., identifying cats vs. dogs in images) – this course tackles the more challenging task of creation. We’ll explore how to build AI models that can generate entirely new content, such as images from text descriptions like “cat” or “dog.”

This course will equip you with a deeper understanding of the core concepts behind modern generative models. We will primarily focus on two foundational architectures:

  • Variational Autoencoders (VAEs): Learn how these powerful models can generate realistic images.
  • Generative Pre-trained Transformers (GPTs): Discover the secrets behind large language models (LLMs) and their ability to generate coherent and contextually relevant text.

Course Structure and Assessment: Learning will be facilitated through a combination of:

  • Lessons: In class assignments.
  • Quizzes: Regular assessments to reinforce your understanding.
  • Homework Assignments: Hands-on practice applying Generative AI techniques.
  • Individual Review Article and Presentation: In place of traditional tests, you will delve deeper into a specific, advanced, area of Generative AI. You will conduct research, write a concise review article, and present your findings to the class. This will foster independent learning and communication skills crucial for staying at the forefront of this rapidly evolving field.

Course Outline:

  1. Introduction to Generative AI:
    • Distinguishing between Discriminative and Generative AI
    • Exploring Unsupervised Machine Learning and its role in generation
    • Understanding the Manifold Hypothesis and its implications
    • Reviewing essential concepts from Deep Learning and Probability Theory
  2. Variational Autoencoders (VAEs): Deep dive into the architecture and applications of VAEs for image generation.
  3. Generative Pre-trained Transformers (GPTs) and Large Language Models (LLMs): Unraveling the power of GPTs and LLMs for text generation, understanding their architecture, and exploring their capabilities. (You will train a GPT2 model from scratch.)
  4. Advanced Topics in Generative AI: (Individual Presentations)
    • Prompt Engineering and Optimization
    • Fine-Tuning LLMs
    • Diffusion Models
    • AI Agents
    • Retrieval Augmented Generation (RAG)

Why Take This Course?

Generative AI is rapidly transforming numerous industries, from art and design to entertainment and software development. This course provides you with the core knowledge and some practical skills to take part in this exciting revolution. You’ll gain hands-on experience with cutting-edge techniques, preparing you for research or industry roles at the forefront of AI innovation.