Conditional Score

Conditional score-based models are generative models that learn to estimate the gradient of a data distribution's probability density function, conditioned on some input information. Current research focuses on improving these models' accuracy and applicability across diverse data types, including images, time series, and medical data, often employing architectures based on diffusion processes and optimal transport. These advancements enable improved data generation, particularly in scenarios with incomplete or unpaired data, leading to applications in areas like medical image completion, time-series forecasting, and image translation. The resulting improvements in data generation and inference have significant implications for various scientific fields and practical applications.