Diffusion Models: Generating Data by Reversing Noise
Think of diffusion models as learning to reverse a "random walk." They take a clean data point, gradually add noise until it's unrecognizable, and then train a model to reverse that process step-by-step. This allows them to start with pure noise and guide it back into a coherent sample that resembles the original dataset. The footgun is that this multi-step reversal makes generation computationally intensive compared to single-pass models.
Diffusion models generate new data by learning to reverse a "random walk." Imagine taking a data point and gradually adding random noise over many steps until it's pure static. The model learns the statistical path to reverse this degradation. For generation, it starts with pure noise and iteratively applies this learned reversal process, guiding the noise back into a coherent sample that matches the original data's distribution. The key footgun is that this iterative, multi-step sampling is inherently slow, making generation less efficient than other methods.
Read the original → Wikipedia: Diffusion model
- #generative ai
- #machine learning
- #latent variable model
Get five bites like this every day.
Tezvyn delivers a daily feed of 60-second tech bites with quizzes to lock in what you learn.