Search papers, labs, and topics across Lattice.
This paper rigorously analyzes discrete diffusion models by deriving the continuous-time Markov chain (CTMC) evidence lower bound (ELBO) for various noising processes, establishing the Oracle Distance theorem which equates the negative ELBO to data entropy plus the path KL divergence. It reveals that the unique optimizer corresponds to the conditional expectation of the true reverse jump rate, leading to a unified understanding of different loss functions in the literature, including MDM, UDM, SEDD, and GIDD. The findings demonstrate that all noising processes share the same optimal negative ELBO, providing insights into the relationships between denoiser, cavity, and score methods in diffusion modeling.
The Oracle Distance theorem reveals that all noising processes achieve the same optimal negative ELBO, linking diverse loss functions in diffusion models.
What does a discrete diffusion model learn: a denoiser, a score ratio, or a bridge plug-in predictor? At the level of jump rates, these are one object in different coordinates, and reading a neural network in the wrong coordinate changes the process being trained and sampled. Starting with a rigorous derivation of the continuous-time Markov chain (CTMC) ELBO for any noising process, boundary terms included, we prove the \emph{Oracle Distance} theorem: the negative ELBO is exactly equal to the data entropy plus the path KL from the oracle reverse process to the learned one, not merely a bound. Its unique optimizer is therefore the conditional expectation of the true reverse jump rate given the current noisy state, and its irreducible cost is the rate at which the forward process $Z_t$ destroys information about the clean data $Z_0$, $-\tfrac{d}{dt}I(Z_0; Z_t)$, so every noising process shares the same best achievable negative ELBO: the data entropy. For sequences with token-factorizing noise, the oracle projection yields three exact coordinates for the optimizer: denoiser, cavity (bridge plug-in), and score, with closed-form conversions among them. This framework identifies which law each loss in the literature actually optimizes, recovering MDM, UDM, SEDD, and GIDD as special cases; explains why denoiser and cavity coincide for masked diffusion but not for uniform diffusion; proves that a denoiser parameterization makes the uniform ELBO diverge at initialization while the bridge plug-in stays finite; and calibrates ELBO implementations exactly at initialization. Every identity is verified numerically, without approximation, on an exactly solvable model.
ETH