Empirical Analysis of Decoding Biases in Masked Diffusion Models (2026.acl-long)
Copied to clipboard
Pengcheng Huang, Tianming Liu, Zhenghao Liu, Yukun Yan, Shuo Wang, Tong Xiao, Zulong Chen, Maosong Sun
| Challenge: | Existing MDMs employ uncertainty-based decoding strategies that limit their reasoning ability and ultimately degrade generation quality. |
| Approach: | They propose a framework that regularizes uncertainty-based decoding by incorporating two complementary priors to shape global decoding trajectories and promote content informativeness. |
| Outcome: | The proposed framework outperforms existing decoding strategies by more than 7% while achieving comparable performance to autoregressive models of similar parameter scales. |
Similar Papers
DecoCal: Decoding with Calibration in Diffusion Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Diffusion Large Language Models (DLLMs) generate text via iterative token denoising . but decoding is challenging, with many tokens appearing predictable early . |
| Approach: | They propose a Decoding framework that performs Calibration of token-level confidence across diffusion steps and leverages the calibrated results to guide decoding decisions. |
| Outcome: | Experiments on multiple DLLMs and benchmarks show that DecoCal improves generation accuracy compared to existing strategies. |
RACC: Regret-Aware Confidence Calibration for Consistent Masked Discrete Diffusion Decoding (2026.findings-acl)
Copied to clipboard
| Challenge: | Masked Discrete Diffusion Models (MDMs) enable parallel generation via iterative refinement, but their current decoding paradigms are static and myopic. |
| Approach: | They propose a Regret-Aware Confidence Calibration framework that aligns decoding decisions with the model’s latent self-correction capabilities. |
| Outcome: | The proposed framework aligns decoding decisions with model’s latent self-correction capabilities. |
Reward-Weighted Sampling: Enhancing Non-Autoregressive Characteristics in Masked Diffusion LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Masked diffusion models (MDMs) are promising non-autoregressive alternatives for large language modeling. |
| Approach: | They propose a method that leverages an external reward model to provide a principled global signal during the iterative diffusion process. |
| Outcome: | The proposed method improves non-autoregressive generation orders and performance across evaluation metrics. |
Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow (2026.findings-acl)
Copied to clipboard
Yangyang Zhong, Yanmei Gu, Zhengqing Zang, Xiaomeng Li, Yuqi Ding, Xibei Jia, Yuting Shen, Zhenzhong Lan, Liwang Zhu, Weiping Liu, Junlin Zhou, Haisheng Liu, Zhong Xin Yu, Pengxin Luo, Donglian Qi, Yunfeng Yan, Junbo Zhao
| Challenge: | Autoregressive (AR) language models dominate modern natural language processing due to strong likelihood-based training objectives and reliable left-to-right decoding. |
| Approach: | They characterize MDLM behavior along two dimensions: parallelism strength and generation order . authors propose a Generate-then-Edit paradigm that mitigates dependency loss . |
| Outcome: | The proposed model improves on tasks that require "backward information" the Generate-then-Edit paradigm improves parallel decoding efficiency while reducing dependency loss. |
DOS: Dependency-Oriented Sampler for Masked Diffusion Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing decoding strategies for pre-trained MDLMs rely on token-level uncertainty criteria, while largely overlooking sequence-level information and inter-token dependencies. |
| Approach: | They propose a training-free decoding strategy that leverages inter-token dependencies to inform token updates during generation. |
| Outcome: | Empirical results show that the proposed approach consistently achieves superior performance on both code generation and mathematical reasoning tasks. |
PURE: Post-hoc Unlocking and REfinement for Discrete Diffusion Decoding (2026.findings-acl)
Copied to clipboard
| Challenge: | Masked diffusion language models (MDLMs) are limited by a monotonic unmasking policy, where committed tokens cannot be revised. |
| Approach: | They propose a training-free inference algorithm for two-phase decoding that unlocks unstable regions through deterministic window masking and stochastic leftward relaxation. |
| Outcome: | The proposed algorithm significantly improves accuracy on reasoning benchmarks on GSM8K. |
You Can Have a Second Chance: Unbiased and Multi-bit Watermarking for Diffusion Language Models with Regret-based Remasking (2026.acl-long)
Copied to clipboard
| Challenge: | Existing sequential LLMs cannot be directly applied to DLMs, as their generation order is arbitrary. |
| Approach: | They propose a stability-aware constraint that allows watermarking only in stable contexts and a bit-controlled, unbiased modulation to preserve the original DLM output distribution. |
| Outcome: | The proposed scheme achieves stable watermarking with minimal quality impact while maintaining high detection accuracy and multi-bit capacity. |
T⋆: Progressive Block Scaling for Masked Diffusion Language Models Through Trajectory Aware Reinforcement Learning (2026.acl-short)
Copied to clipboard
| Challenge: | Autoregressive (AR) modeling via next-token prediction dominates scaling practice and deployed systems. |
| Approach: | They propose a TraceRL-based curriculum for progressive block-size scaling in masked diffusion language models. |
| Outcome: | The proposed curriculum outperforms direct large-block TraceRL on two SDAR scales and three benchmarks and retains block-size-specific non-monotone updates while improving accuracy. |
Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for full-attention dLLMs rely on random masking strategies that overlook intrinsic token dependencies. |
| Approach: | They propose an attention-guided denoising and optimization framework that aligns training and optimization with attention-derived dependencies. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on mathematical and coding benchmarks. |
Beyond Hard Masks: Progressive Token Evolution for Diffusion Language Models (2026.acl-long)
Copied to clipboard
Linhao Zhong, Linyu Wu, Bozhen Fang, Tianjian Feng, Chenchen Jing, Wen Wang, Jiaheng Zhang, Hao Chen, Chunhua Shen
| Challenge: | Existing Diffusion Language Models rely on hard binary masking and discrete token assignments, which hinder the revision of early decisions. |
| Approach: | They propose a diffusion-based language modeling approach that replaces hard binary masks with evolving soft token distributions. |
| Outcome: | The proposed approach outperforms existing DLMs on multiple benchmarks. |