Home Glossary Item Diffusion
« Back to Glossary Index

Diffusion in the context of artificial intelligence (AI) can take multiple forms. One common usage is in the realm of diffusion models, which are a class of generative models used in machine learning. These models create new data instances that resemble the training data. For instance, a diffusion model trained on a dataset of images could generate similar, yet distinct, images. These models work by starting from a random initial state and gradually ‘diffusing’ towards states that resemble the training data.


Diffusion models are highly beneficial in various applications within AI. They’re used to create realistic synthetic data for training other machine learning models, especially in scenarios where actual data is scarce or sensitive. They can also be used in data augmentation, image denoising, super-resolution, and inpainting tasks. These models bring randomness into the process of generating new data, which can encourage diversity and creativity.

The term ‘diffusion’ can also be used to describe the spread or adoption of AI technology itself. This includes how AI innovations spread across different sectors, organizations, and geographical regions. The speed and pattern of this diffusion can be influenced by a variety of factors, ranging from governmental policies and public acceptance, to infrastructural readiness and the perceived utility of the AI technology itself. Much like in a scientific context, this diffusion also tends to move from areas of higher concentration (early adopters or AI-forward companies and countries) to areas of lower concentration.

« Back to Glossary Index