Classifier-Free Guidance

Guided sampling (Classifier-Free Guidance, CFG for short) is a technique proposed by OpenAI in 2022 for enhancing sample quality in generative models (especially in the fields of image generation and natural language generation). It aims to improve the control ability of generative models to make them more consistent with specific goals or conditions while avoiding the use of additional classifiers for conditional constraints. In this way, the generative model can generate samples that are highly relevant to the given conditions without explicit conditional classifiers.