Anari
A researcher who in 2026 proposed a method using rejection sampling to correct the parallel sampling bias of masked diffusion models, decomposing the log acceptance probability into a sum of token-level terms.
Timeline (last 90 days)
Anari et al. (2026) proposed using rejection sampling to correct the distributional bias in parallel decoding of masked diffusion models, decomposing the log acceptance probability into a sum of token-level terms.
Anari et al.'s scheme requires one score query for every token unmasked, so the number of queries grows linearly with the block size.
All Facts (2)
Anari et al. (2026) proposed using rejection sampling to correct the distributional bias in parallel decoding of masked diffusion models, decomposing the log acceptance probability into a sum of token-level terms.
50%UnverifiedAnari et al.'s scheme requires one score query for every token unmasked, so the number of queries grows linearly with the block size.
50%