The Role of Differential Privacy in GDPR Compliance
The Role of Differential Privacy in GDPR Compliance
复制标题
差异隐私在 GDPR 合规性中的作用
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
D. Desai
中科院分区：
文献类型：
--
作者：
Rachel Cummings;D. Desai
欧盟《通用数据保护条例》（GDPR）赋予个人控制公司所持有的个人数据的删除的权利。GDPR还允许企业保留匿名汇总数据和统计结果。不幸的是，大多数推荐系统（以及许多其他类型的机器学习模型）在训练时会记住单个数据条目，因此没有充分匿名以符合GDPR。差分隐私在形式上防止了记忆化和其他类型的过拟合，并且还允许在各种机器学习任务中进行准确的分析。在这份立场文件中，我们主张差异化私人学习应该是GDPR推荐系统的首选方法。
The EU General Data Protection Regulation (GDPR) empowers individuals with the right to control erasure of their personal data held by firms. GDPR also allows firms to retain anonymized aggregate data and statistical results. Unfortunately, most recommender systems (and many other types of machine learning models) mem-oize individual data entries as they are trained, and thus are not sufficiently anonymized to be GDPR compliant. Differential privacy formally prevents against memoization and other types of overfitting, and additionally allows for accurate analysis in a wide variety of machine learning tasks. In this position paper, we advocate that differentially private learning should be the preferred method for GDPR-compliant recommender systems.