PulseAugur
中
实时 08:49:48
English(EN) Reward-Driven Learning under Prompt-Level Differential Privacy

新的 RLVR 方法为语言模型训练提供差分隐私保护

研究人员开发了一种新颖的方法,在满足提示级别差分隐私的同时,使用可验证奖励(RLVR)的强化学习来训练语言模型。该方法确保发布的模型权重对于单个训练问题是差分私有的。该方法聚合梯度、裁剪贡献、添加高斯噪声并组合隐私损失,提供了已知的首个 RLVR 训练的差分隐私保证。使用 Qwen2.5-1.5B-Instruct 进行的实验表明,即使在隐私限制下,奖励信号也能显著提高 MATH 和 GSM8K 等数学任务的准确性,其性能优于有监督微调方法,并保留了非私有方法的大部分收益。 AI

影响 这项研究有可能在不显著降低性能的情况下,实现更私有的 AI 模型开发,特别适用于敏感数据应用。

排序理由 该集群包含一篇学术论文,详细介绍了一种具有差分隐私保证的语言模型训练新方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

新的 RLVR 方法为语言模型训练提供差分隐私保护

本文如何被排名

Signal score
15 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇学术论文,详细介绍了一种具有差分隐私保证的语言模型训练新方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety, model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.LG TIER_1 English(EN) · Jiachen Zhao, Antonia Januszewicz, Taeho Jung ·

    差分隐私下的奖励驱动学习

    arXiv:2610.07212v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) trains a language model on problems that may themselves be confidential, and the trained model can reveal which problems it saw. We study RLVR under prompt-level differential pri…