Lecture 2 - Amr El Abbadi: Towards Scalable and Practical Privacy Preserving Information Retrieval

This is the second lecture of the tutorial. You can also attend online in Zoom.

The first lecture will be held on Friday 19 April at 16:00-18:00 in lecture room T4, CS building & online in Zoom

Towards Scalable and Practical Privacy Preserving Information Retrieval

Amr El Abbadi
Professor of Computer Science
University of California at Santa Barbara

Our goal in this tutorial is to demonstrate how private access of data can become a practical reality in the near future. Our focus is on supporting oblivious queries and thus hide any associated access patterns on both private and public data. The tutorial compromises of two lectures. To get the most out of the lectures, please try to attend both in sequence as the second lecture builds on material covered in the first lecture.  

The main focus of the tutorial will be on privately accessing public data.  PIR (Private Information Retrieval) is the main mechanism proposed in recent years for private access of public data.  However, current PIR proposals are inefficient especially with large data sets.  In the first lecture, we will motivate the problem by discussing a novel efficient PIR data structure to design a scalable infrastructure for voice communication that will hide meta-information. We will cover the basic concepts of PIR such as its types, construction, and critical building blocks, including homomorphic encryption. The solution to this challenge motivates and has significant ramifications on diverse data management problems such as designing scalable systems for oblivious search for documents from public repositories as well as private query processing over public databases. 

Current PIR proposals usually require the server to consider data as an array of elements and clients retrieve data using an index into the array. This latter restriction limits the use of PIR in many practical settings, especially for key-value stores, where the client may be interested in a particular key, but does not know the exact location of the data at the server. The second lecture will discuss recent efforts to overcome these limitations, using Fully Homomorphic Encryption (FHE).  Focus will be on how to improve the performance, scalability and expressiveness of privacy preserving queries on public data to support both private key (not index) based retrieval as well as private top-k qualifying documents corresponding to a given keyword search query.  This will involve techniques to make efficient fundamental operations, like equality and matrix multiplication.

The presentation will be based on the following recent references.

Ishtiyaque Ahmad, Yuntian Yang, Divyakant Agrawal, Amr El Abbadi, Trinabh Gupta: Addra: Metadata-private voice communication over fully untrusted infrastructure. OSDI 2021

Ishtiyaque Ahmad, Laboni Sarker, Divyakant Agrawal, Amr El Abbadi, Trinabh Gupta: Coeus: A System for Oblivious Document Ranking and Retrieval. SOSP 2021: 672-690

Ishtiyaque Ahmad, Divyakant Agrawal, Amr El Abbadi, Trinabh Gupta:
Pantheon: Private Retrieval from Public Key-Value Store. Proc. VLDB Endow. 16(4): 643-656 (2022)