Diffusion models are a key part of not just image generation, but also medical diagnostic systems
Diffusion models exploded onto the world stage a mere two years ago. The technology had been around for a while, but it was only when we all experienced the revolution of AI image generation that it came into stark focus.
Suddenly everyone and their dog could create astounding works of art simply by typing a prompt into a text box.
But diffusion models are not just about art and image creation. The scientific world, music and even Hollywood, have started to to understand the benefits of this powerful AI technology.How does it work?
Diffusion works by adding noise to training data until it becomes completely unrecognizable, and then reversing the process to create novel data from the original learnings.
It’s rather like a sculptor ‘uncovering’ their vision by slowly chipping away at the stone wood or plaster in front of them.
The clever bit is teaching the AI model to understand how to recreate a new version of the original data by subtracting noise until the desired result comes into ‘focus’.
Diffusion is one of the hardest AI concepts to explain to the layperson. When people talk about AI stealing art or music, what they don’t realize is the diffusion models which deliver these miraculous results, are not acting directly on the original training data, but are using that data as a starting point from which to create something completely new and unique.
