Score Based

Score-based methods leverage the gradient of a probability distribution (the "score") to solve various problems across diverse fields. Current research focuses on applying score-based approaches in generative modeling, Bayesian inference (particularly for inverse problems in imaging and other areas), and adversarial attacks/defenses, often employing diffusion models, autoencoders, and neural networks to estimate or utilize the score function. These methods offer advantages in scalability and efficiency for tasks like image reconstruction, data selection, and automated scoring, impacting fields ranging from astronomy and medical imaging to machine learning and education.