Transformer-Based Masked Diffusion Models Provably Learn Distributions with Latent Structure XXV36V6CK8

URI	`m2://res/XXV36V6CK8`
Kind	preprint
Creators	Jingda Wu, Changxiao Cai
Date	2026-10-08
Subjects	arxiv cs.LG, arxiv cs.IT, arxiv math.IT, arxiv math.ST, arxiv stat.ML, arxiv stat.TH
Scope	border

Expressions

e1: version, preprint

Date	2026-10-08
Ids	arXiv 2610.12544

Items in this resource

The mentions of the inventory that name this resource. Each name links to the card of its inventory row.

Item	Kind	Where	Source
Estimation error for shallow tree	statement	theorem 3	arxiv
3KM381W9PV	statement	lemma 9	arxiv
Approximation under low R\'enyi entropy	statement	lemma 2	arxiv
Sampling guarantees under shallow tree structure	statement	theorem 4	arxiv
Multiplication on a bounded rectangle	statement	lemma 13	arxiv
TC and DTC schedules	statement	lemma 18	arxiv
Uniform approximation of the exponential function (Lemma F.12 in the arXiv version	statement	lemma 14	arxiv
Dependence trees and tree distributions	definition	definition 1	arxiv
Tree-mixture approximation	statement	proposition 2	arxiv
Global softmax stability	statement	lemma 16	arxiv
$\delta$-regularization	definition	definition 4	arxiv
Single-tree parallel construction	statement	lemma 8	arxiv
Mixture of hidden Markov trees	definition	definition 2	arxiv
Uniform approximation of logarithm on the interval	statement	lemma 15	arxiv
Single-tree message construction	statement	lemma 10	arxiv
Exact realization of clipping	statement	lemma 11	arxiv
Decomposition of the approximation error	statement	lemma 7	arxiv
Statistical error	statement	proposition 1	arxiv
Localized oracle inequality for a logit class	statement	proposition 3	arxiv
Localized bounds from a uniform cover	statement	lemma 4	arxiv
Per-token denoising loss is average conditional KL	statement	lemma 5	arxiv
KL approximation error	definition	definition 3	arxiv
Capacity of the sparse transformer class	statement	lemma 3	arxiv
SCSSRRTFQS	statement	lemma 1	arxiv
Estimation error of mask predictor	statement	theorem 1	arxiv
Height-dependent approximation of regularized mixture logits	statement	proposition 5	arxiv
End-to-end sampling error of MDM	statement	theorem 2	arxiv
Uniform approximation of regularized mixture logits	statement	proposition 4	arxiv
Properties of $\delta$-regularization	statement	lemma 6	arxiv

Catalogue records

arxiv: 2610.12544 (work)