5 min read
Diffusion
Diffusion is a machine learning framework that can learn probability distribution of any Data.
Probability distribution shows the likelihood of each value occurring in a random experiment of process.
Probability refers to the likelihood of a specific outcome in the dice roll, like rolling a 3, which is 1/6. The probability distribution lists all possible outcomes and their probabilities together.
LLMs at its core learn data distribution, what is special about Diffusion?
Diffusion Modals can learn piData (probability distribution) from relatively small amount of data?
Diffusion Process:
Add noise to training image, and then train model to de-noise it. one the model learns to de-noise it, for inference give it a random noisy image and ask to de-noise it to a something.
