Ontology of Diffusion Models: Tools, Language and Architecture Design
Matias del Campo · 2024
Diffusion models are a family of deep generative models that have surfaced in recent years. This chapter provides a vantage point for the ontology of diffusion models. Three versions of diffusion models are being discussed: stochastic differential equations, denoising diffusion probabilistic models, and score-based generative models. To explain all the models, and even to attempt a categorization and discussion of all models regarding their cultural impact would be a profoundly naive approach to the problem, thus in a case diffusion models are representative for ideas about language, culture, and tools. In its very core humans are cultural beings, despite the fact that humans are generally measured by its ability to create artefacts. Just to mention a few of the various challenges for learning tools: facial recognition, fraud detection, medical diagnostics, voice recognition, traffic optimization, shipping automation, and many more.