Learning objectives
Describe the scenario of the 2027 case study, a design studio choosing generative models for commercial image creation, and identify the technical and ethical challenges it raises
Explain how a diffusion model creates an image by learning to reverse a gradual noising process, and why iterative denoising makes generation demanding in processing time, energy and hardware
Discuss the intellectual-property, consent and bias problems that arise when generative models are trained on very large collections of images, and propose policies a studio could adopt
(HL) Explain how a generative adversarial network trains a generator against a discriminator, and describe the effects of an unbalanced pair such as mode collapse and unstable training
(HL) Evaluate combining variational autoencoders, GANs, flow-based models and diffusion models for the requirements of a studio on image quality, speed, cost and control
Continue with a revision plan
These learning objectives are available to everyone. Sign in with an eligible plan to read the full note and worked examples.