Guided Generation

Guided generation focuses on controlling the output of generative models, steering them towards desired properties or constraints. Current research emphasizes leveraging diffusion models and other architectures to generate diverse and nuanced outputs, including minority samples in datasets and explanations for complex models like Graph Neural Networks. This ability to guide generation is crucial for improving the reliability and interpretability of AI systems, impacting fields ranging from data augmentation to explainable AI and creative content generation. Furthermore, efficient methods for guiding large language models are being developed to enhance control and ensure adherence to specific structural or semantic rules.