PulseAugur
实时 22:28:12
English(EN) Discrete Stochastic Localization for Non-autoregressive Generation

新的DSL框架增强了非自回归生成模型

研究人员引入了离散随机定位(DSL),一种新的连续状态非自回归生成框架。该方法旨在通过提供一种对信噪比不变的更灵活的表示来改进现有的离散扩散模型。在OpenWebText数据集上,使用DSL对预训练模型进行微调已显示出分布忠实度的显著提高,甚至支持以更少的步数进行更快采样。 AI

影响 引入了一种新颖的框架,提高了生成模型的分布忠实度和采样效率。

排序理由 该集群包含一篇详细介绍非自回归生成新方法的学术论文。

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

报道来源 [2]

  1. arXiv cs.CL TIER_1 English(EN) · Yunshu Wu, Jiayi Cheng, Longxuan Yu, Partha Thakuria, Rob Brekelmans, Evangelos E. Papalexakis, Greg Ver Steeg ·

    Discrete Stochastic Localization for Non-autoregressive Generation

    arXiv:2602.16169v2 Announce Type: replace-cross Abstract: Continuous diffusion is a natural framework for non-autoregressive generation but has generally lagged behind masked discrete diffusion models (MDMs) on discrete sequence generation. We argue that the bottleneck is not con…

  2. arXiv cs.LG TIER_1 English(EN) · Yunshu Wu, Jiayi Cheng, Longxuan Yu, Partha Thakuria, Rob Brekelmans, Evangelos E. Papalexakis, Greg Ver Steeg ·

    Discrete Stochastic Localization for Non-autoregressive Generation

    arXiv:2605.12836v2 Announce Type: replace Abstract: Continuous diffusion is a natural framework for non-autoregressive generation but has generally lagged behind masked discrete diffusion models (MDMs) on discrete sequence generation. We argue that the bottleneck is not continuit…