AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Learning Unmasking Policies for Diffusion Language Models

Metod Jazbec, Theo X. Olausson, Louis Béthune, Pierre Ablin, Michael Kirchhof, Joao Monteiro, Victor Guilherme Turrisi da Costa, Jason Ramapuram, marco cuturi

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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
use a multiplicative correctness-times-speed reward instead of an additive penalty (ablation)RL unmasking policy (alpha=1) · GSM8K training on LLaDAtraining stability / reward hackingImproved
train a policy for full diffusion (BL=256) instead of using confidence heuristicsLLaDA-8B-Instruct sampling · GSM8K, ~12 NFEsaccuracy (%): 30% -> 50%Improved
add expert steering using Fast-dLLM semi-AR samples to full-diffusion RL training (ablation)RL unmasking policy (BL=256) · GSM8K and MATH-500, mid-to-high NFEsaccuracy: approx. 80% GSM8K and 35% MATHImproved
add expert steering to RL training (ablation)RL unmasking policy · full diffusion trainingtraining stability / controllability via alphaHurt
use RL policy instead of Fast-dLLM in semi-AR mid-to-high NFE rangeLLaDA-8B-Instruct · GSM8K and MATH-500, BL=32accuracy vs NFEsNo clear effect
scale policy unmasking probabilities by beta at test time (ablation)RL policy trained with alpha=1 · MATH-500 semi-AR, ~25 NFEsaccuracy: 10% -> 20% vs Fast-dLLMImproved
use DPLS (dynamic Plackett-Luce) policy likelihood instead of Bernoulli (ablation)RL unmasking policy · GSM8K, LLaDAaccuracy vs NFEsNo clear effect
feed top-50 confidences instead of max confidence per position (ablation)RL unmasking policy · GSM8K, LLaDAaccuracy vs NFEsHurt
use dLLM hidden states as policy input (300M-parameter head) instead of confidences (ablation)RL unmasking policy · GSM8K, LLaDAaccuracy and training stabilityHurt
zero out mask vector or time input at test time (ablation)RL policy (alpha=1) · GSM8K and MATH-500, BL=32accuracyHurt
evaluate LLaDA-trained policies on Dream without retrainingRL unmasking policy · GSM8K, Dream-7Baccuracy vs NFEsMixed
transfer math-trained policies to coding without retrainingRL unmasking policy · HumanEval and MBPPaccuracyHurt

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