Soft Regularization

A method that approximately satisfies constraints by encouraging generated results to approach a target within the loss function. In topology-aware generative models it cannot guarantee an exact match to the target persistence diagram and yields only approximately similar results.