Generative AI: Introduction to Diffusion Models for Text Generation

In this course, Wuraola Oyewusi—an experienced data scientist and machine learning and artificial intelligence professional—introduces you to diffusion models, which are popular with image generation, and how you can apply these models in natural language generation tasks. Explore how diffusion models work as an alternative to traditional autoregressive approaches and develop your understanding of their unique advantages in generating diverse and controllable text outputs. Through hands-on exercises, you'll build and fine-tune text diffusion models, work with popular implementations on platforms like Hugging Face, and gain practical experience applying these cutting-edge techniques to real-world text generation challenges. This course prepares you to use diffusion models effectively for generating the text you need.

Learning objectives
Understand the fundamentals of Generative AI and its specific applications in Natural Language Generation (NLG), including identifying various subtasks within text generation.
Compare and contrast Diffusion Models with Autoregressive Models for text generation, clearly articulating the distinct advantages and challenges of each approach.
Build a basic text diffusion model from scratch, utilizing pre-trained BERT embeddings and understanding the essential roles of noise addition, denoising, and text decoding.
Fine-tune a diffusion model to perform conditional text generation, enabling the model to produce text aligned with specific prompts or contextual inputs.
Explore and interact with advanced diffusion models, such as Google Gemini Diffusion Model, within platforms like AI Studio.