cs.CLNov 7, 2026

FEROM: Frontier Endogenous Reveal-Order Marginal Policy Optimization for Masked Diffusion LMs

Authors: Zian Su, Ziyang Huang, Kaiyuan Zhang, Xiangyu Zhang

Organizations: Purdue University · Johns Hopkins University · Rutgers University

Abstract

Masked diffusion language models generate text by iteratively unmasking positions, with practical samplers often selecting reveal positions from the model's own logits. The reveal order is therefore an endogenous latent variable of the rollout process with an enormous discrete latent space. Policy optimization objectives relying on pre-defined schedules or sampled trajectories introduce either bias or high variance. In this paper, we propose FEROM, Frontier Endogenous Reveal-Order Marginal Policy Optimization, which approximates policy gradients for the rollout-induced marginal response policy of masked diffusion LMs. FEROM derives a Rao-Blackwellized policy-gradient identity over latent reveal orders and expresses the resulting estimator as posterior edge occupancy on a reveal-state DAG. To make marginalization practical, we introduce Frontier Reveal Marginalization, a budgeted estimator that approximates this marginal edge structure by combining CDF-based computation of local probabilities of the endogenous reveal process with frontier expansion of high-mass partial states. Integrated into a GRPO-style objective, FEROM replaces single-path log-scores with a detached token-score surrogate over posterior-relevant reveal edges. Experiments on math and coding tasks show comparable or improved results over existing methods under matched scoring-state budgets. An offline proxy study further shows improved marginal-gradient approximation with increased scoring budgets.

Figures & tables

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Jul 16, 2026

Mask-Aware Policy Gradients for Diffusion Language Models

Reinforcement learning has proven effective for improving reasoning in large language models, but extending it to Masked Diffusion Language Models (MDLMs) remains challenging due to the intractability of the log-likelihood estimation. Existing approaches approximate this log-likelihood by modeling only the token predictions, ignoring the order in which positions are unmasked during generation. We observe that MDLM generation involves two decisions at each step: what tokens to place at each masked position and which positions to remask. We formalize this as a two-stage action MDP, showing that the policy gradient naturally decomposes into a token term and a masking term. Combining optimization of both terms leads to state-of-the-art outcomes on mathematical reasoning and coding benchmarks, with scores of 87.1% on GSM8K and 53.4% on MBPP.
Oct 22, 2026

What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models

Masked diffusion language models generate tokens in parallel, but their efficiency depends on which predictions are committed. Learned unmasking policies improve the speed--accuracy trade-off, yet differ in supervision, inputs, and capacity, obscuring which choices matter. We view supervised policies as distilling a decoding procedure through a teacher and examine independently what the teacher labels as committable, on which states the policy imitates it, what the policy observes, and how much the number of function evaluations (NFE) must decrease to offset policy overhead. We find that labeling predictions matching the final output is overly permissive at low NFE, and hidden states are weak alone but can complement confidence. About a 10% NFE reduction suffices to offset policy overhead for policies up to 7.5M parameters across tested settings. We therefore train a policy on trajectory-contraction targets, which accept only commitments staying on a reference trajectory, using confidence and hidden states. On LLaDA-8B-Instruct, the resulting 856K-parameter policy achieves the best accuracy--NFE frontier and highest throughput on GSM8K, is competitive at moderate targets on MATH-500 and HumanEval, and largely transfers to Dream-7B-Instruct.
Nov 17, 2026

ReMaRL: Learning to Unmask and Remask in Frozen Diffusion Language Models

Masked diffusion language models generate tokens in parallel, and the order in which tokens are unmasked affects final-answer accuracy. Several methods learn unmasking policies from task rewards, but these policies cannot remask earlier predictions. Remasking lets the model correct those predictions using newly generated context. Heuristic rules guide both unmasking and remasking, while supervised remaskers learn to detect token errors. Neither approach directly optimizes these decisions for final-answer correctness. To address this gap, we introduce ReMaRL (Remasking with Reinforcement Learning), a lightweight policy that learns which tokens to unmask, retain, or remask. We train this roughly 0.3M-parameter policy with Group Relative Policy Optimization (GRPO) from final-answer rewards while keeping the diffusion model frozen. With LLaDA-8B-Instruct, ReMaRL achieves the highest scores among the compared unmasking and remasking policies within the tested inference budgets of 4-32 model calls on math and 8-16 calls on code.