Adversarial Masked Diffusion for Path-Generation Decoupling

Diffusion models usually learn generation and reverse-path construction through the same regression objective. This coupling is stable, but it also makes generation a conditional point-estimation problem: when a corrupted state admits many plausible clean completions, full-sample regression tends to favor averaged solutions. We propose Adversarial Masked Diffusion (AMD), a masked diffusion framework for path-generation decoupling. In AMD, generation and path formation are assigned to different learning signals: clean-sample generation is learned through adversarial distribution matching, while reverse-path preservation is enforced only through regression on the coordinates that remain visible in the corrupted input. The discriminator compares real clean samples with generated clean predictions without observing the corrupted state, mask, or conditioning image. The masked diffusion process uses independent Bernoulli masks in data space: every corruption state is drawn with a fresh coordinate mask and fresh replacement values. During sampling, every lower-corruption state is likewise formed using a newly sampled mask and replacement noise at the target level. A visible-coordinate path loss anchors the prediction to the unmasked part of the input, preventing the adversarial generator from ignoring the reverse state. Thus, AMD decouples generation from path preservation: adversarial learning matches the clean data distribution, and regression constructs the reverse path. As an independent training optimization, partial regularization evaluates the discriminator gradient regularizer on a fixed-size subset of every minibatch while retaining adversarial training on the full batch.