AI Model Architectures

Generative AI uses several architecture families. Transformers model sequences, diffusion and flow models synthesize through iterative transformations, GANs train competing networks and VAEs learn probabilistic latent spaces.

Platforms like Midjourney, Dalle-2, Stable Diffusion, and ChatGPT have increased popularity of Generative AI, leading to increased interest in multi-headed and single-headed models, as well as graph models. Multi-modal graph models can handle graph-structured data that includes text, photos, videos, and audio. Graph neural network models can be single-headed or multi-headed, can perform tasks such as node classification and connection prediction, and can use convolutional networks, attention networks, and generative graph models.