AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Stop Training for the Worst: Progressive Unmasking Accelerates Masked Diffusion Training

Jaeyeon Kim, Jonathan Geuter, David Alvarez-Melis, Sham M. Kakade, Sitan Chen

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Sparse concepts and hypothesis testing · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace random masking with PUMA progressive unmasking forward process125M MDM from scratch · TinyGSM, GSM8K test accuracyiterations to reach accuracy: 2.3x fasterImproved
Add PUMA on top of autoregressive initialization125M MDM (ARM init) · TinyGSM, GSM8K test accuracyiterations to reach accuracy: approximately 4.0x speedupImproved
Add PUMA to block diffusion training (block size 256)Block diffusion 125M · TinyGSM, GSM8K test accuracyaccuracy vs iterationsImproved
Add PUMA to LoRA fine-tuning of Dream-Coder-Base-7BDream-Coder-Base-7B · HumanEvalPass@1: about +10 percentage points (vanilla at most +3%)Improved
Add PUMA to LoRA fine-tuning of Dream-Coder-Base-7BDream-Coder-Base-7B · MBPPPass@1: approximately +5 percentage pointsImproved
Add PUMA on Sudoku6.8M bidirectional MDM · Sudokutraining speedup: 1.7xImproved
Add K-scheduling (K=42 down to 12) rather than fixed K=12 (ablation)PUMA 125M · TinyGSM, GSM8K test accuracytraining speedupImproved
Lower confidence threshold to 0.4 or 0.65 (default 0.9) (ablation)PUMA 125M · TinyGSM, GSM8K test accuracytraining efficiencyHurt
Vary confidence threshold among 0.75 to 0.95 (ablation)PUMA 125M · TinyGSM, GSM8K test accuracytraining efficiencyNo clear effect
Change inference unmasking policy (Top-K margin, entropy, etc.) for a PUMA-trained model (ablation)PUMA 125M · TinyGSM, GSM8K test accuracyaccuracy: 2.2x (Top-K margin), 2.3x (entropy)No clear effect
Add PAPL loss reweighting (ablation)MDM baseline and PUMA · Sudoku and TinyGSMaccuracyHurt
Use PUMA instead of random masking (wall-clock)125M MDM · TinyGSMtraining throughput (it/s): 3.78 -> 3.41Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy