Generative AI

Introduction

This script serves as an introduction to Generative AI and was developed for the elective module “Generative AI,” offered to master’s students of the “Data Science” program at the University of Applied Sciences Kiel. Built using quarto, this resource is designed to provide an accessible overview of key topics and applications in this rapidly evolving field.

While not an exhaustive guide to Generative AI, the script highlights foundational concepts, modern applications, and practical techniques that empower students to engage with and explore the possibilities of these transformative technologies.

Contents and learning objectives

Contents listed in the module database entry:

Open Source Language Models

• Overview of model lists
• Ollama
• Generation of synthetic text as training sets

Agent Systems

• Llamaindex, LangChain & Haystack
• Function calling
• Data analysis

Embeddings and Vector Stores

• Semantic Search
• Retrieval-augmented generation
• Recommendations

AI Image Generators

• Generative Adversarial Networks (GANs)
• Variational Autoencoders / Diffusion Models
• Generative approaches for image dataset augmentation

Fine-Tuning of LLMs and Diffusion Models

• Examples: LoRA, QLoRA, MoRA

---

Learning objectives listed in the module database entry:

Students

• know the fundamentals of generative AI systems.
• know various modern applications of generative AI systems.
• know the theoretical foundations and practical applications of generative AI systems.

Students

• are able to explain and apply various open-source language models.
• are able to implement and utilize agent systems and their functionalities.
• are able to understand and use embeddings and vector stores for semantic search and recommendations.
• are able to explain and practically apply different methods for image generation.
• are able to fine-tune large language models (LLMs) and diffusion models for specific tasks.

Students

• are able to successfully organize teamwork for generative AI projects.
• are able to report and present team solutions for practical project tasks.
• are able to interpret and communicate the approaches in technical and functional terms.

Students

• are able to work professionally in the field of generative AI systems.
• are able to give and accept professional feedback to different topics of generative AI systems.
• are able to select relevant scientific literature about generative AI systems.

---

Schedule:

Course schedule

Number:	CW:	Date:	Title:	Topics:
1	46	12.11.	Getting started with (L)LMs	Language Model Basics
				Choosing open source models
				Basics of using open source models (Huggingface, Ollama, LLM-Studio, Llama.cpp, …)
2	46	13.11.	Prompting	Prompting strategies
				Generation of synthetic texts
3	47	19.11.	Agent basics	Fundamentals of agents and chain-of-thought prompting
				Examples of agent-frameworks (Llamaindex, LangChain & Haystack)
4	47	20.11.	Embedding-based agent-systems	Semantic embeddings and vector stores
				Retrieval augmented and interleaved generation
5	48	26.11.	Function Calling	Code generation and function calling
6	48	27.11.	Agent interaction	LLM as a Judge
				Constitutional AI
7	49	3.12.	AI image generation I	AI image generator basics
				Diffusion Models and Variational Autoencoders
				Multimodal models
8	49	4.12.	AI image generation II	Generative Adversarial Networks (GANs)
				(Generative) approaches for image dataset augmentation
9	50	10.12.	AI image generation III	Using Open Source AI image generation models
				AI image generators in agent systems
10	50	11.12.	Finetuning Basics	Basics of Finetuning strategies
			Rank adaptation	Fundamentals of High and Low-Rank Adaptation of Language and Diffusion Models
				(Q)LoRA fine-tuning using Unsloth
11	51	17.12.	Alginment	Central principles of Model-Alignment
				Reinforcement Learning from Human Feedback (RLHF)
12	51	18.12.	Project presentations
	3	17.1.	Project submission on moodle