Diffusion Models

Three speakers giving three distinct talks on Diffusion Models.

Evelyn J. Boettcher will give a high level talk on what diffusion models .

PhD student Daniel Brignac will do a deep dive on the math, followed up with

Wesley Giles giving a MLOPs example of how diffusion models are used.

What are Diffusion models.

Diffusion models are a type of generative AI model used to create new data, such as images or audio, from random noise:

• Starting point: Begin with random noise.
• Forward process: Gradually add noise to real data samples until they become indistinguishable from random noise.
• Reverse process: Train a neural network to learn how to remove noise step-by-step.
• Generation: To create new data, start with random noise and use the trained model to progressively remove noise, revealing a new, coherent sample.

Key points:

• Diffusion models learn to denoise data
• They work iteratively, making small improvements each step
• The process mimics how information spreads or "diffuses" in nature
• They've shown impressive results in image and audio generation
  The Gem City Machine Learning (ML) group is part of the GemCity TECH family of user communities in Dayton OH.