Differentially Private Learned Indexes

TL;DR

This paper introduces a novel approach to constructing differentially private indexes using learned indexes, significantly reducing storage costs while enabling efficient predicate queries on encrypted data in TEEs.

Contribution

The paper proposes leveraging learned indexes to create more compact differentially private indexes for encrypted databases, improving storage efficiency over existing methods.

Findings

Learned indexes can reduce index size compared to traditional DP indexes.

The proposed method maintains privacy guarantees while improving efficiency.

Experimental results demonstrate practical benefits in encrypted database querying.

Abstract

In this paper, we address the problem of efficiently answering predicate queries on encrypted databases, those secured by Trusted Execution Environments (TEEs), which enable untrusted providers to process encrypted user data without revealing its contents. A common strategy in modern databases to accelerate predicate queries is the use of indexes, which map attribute values (keys) to their corresponding positions in a sorted data array. This allows for fast lookup and retrieval of data subsets that satisfy specific predicates. Unfortunately, indexes cannot be directly applied to encrypted databases due to strong data dependent leakages. Recent approaches apply differential privacy (DP) to construct noisy indexes that enable faster access to encrypted data while maintaining provable privacy guarantees. However, these methods often suffer from large storage costs, with index sizes typically scaling linearly with the key space. To address this challenge, we propose leveraging learned indexes, a trending technique that repurposes machine learning models as indexing structures, to build more compact DP indexes.