EE Seminar: Efficient Privacy Preserving Computation
(The talk will be given in English)
Speaker: Dr. Hayim Shaul
IDC & Haifa University
Wednesday, January 8th, 2020
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering
Efficient Privacy Preserving Computation
Abstract
Privacy preserving multi-party computation enables several parties to collaboratively compute a function over their secret inputs. Recent advances show that privacy preserving arithmetic operations can be done efficiently, however, comparisons remain impossible since they compromise the privacy of the data. This introduces a new computational model: the arithmetic circuit model. Many algorithms that are efficient in comparison model become inefficient or even infeasible in the arithmetic circuit model.
In this talk we'll demonstrate two techniques for designing efficient algorithms in the arithmetic circuit model:
1. We introduce a double-blinded coin toss and show how to use it for an efficient approximation for the k-nearest neighbors classifier.
2. Inspired by von Neumann's work on fault tolerant computation we show how to reduce the noise in a floating-point arithmetic circuit and thus reduce the bit size of the numbers.
Although motivated by privacy preserving algorithms these techniques are interesting outside the crypto world as well. In this talk we will consider the underlying cryptographic schemes as a black box. No prior knowledge in cryptography is needed.
Short Bio
Hayim has done his postdoc in MIT as an applied cryptographer in a robotics lab, and then in IDC and in Uni. of Haifa. His research focuses on practical privacy preserving multi party computation, and especially on fully homomorphic encryption (FHE). Prior to that Hayim completed his PhD in computational geometry under the supervision of Micha Sharir in Tel Aviv University. In the time between his PhD and his postdoc, Hayim spent a few years co-founding and CTO-ing a network optimization company funded by the IFC (world bank).
