Partition Generative Modeling: Masked Modeling Without Masks

ICLROral2026

Authors
Justin Deschenaux, Lan Tran, Caglar Gulcehre
Affiliation
EPFL - EPF Lausanne
Venue
ICLR 2026
Track
Oral

TL;DR

We show that it is possible to train masked generative models without using MASK tokens, resulting in efficiency gains at inference.

Opening excerpt from the authors’ abstract. source

Read the paper

Topics

generative model

← All ICLR 2026 Oral papers · Browse the whole archive