12 Feb 2015 - Changyu Dong

Abstract

In the past a few years, we have seen a dramatic increase in the scale and financial damage caused by cyber attacks. Compounding the problem, changes in computing -- particularly the booming of Cloud computing and collaborative data analysis -- has added another layer of complexity to the security landscape. Traditionally, an organization can lock their data in secure storage and process it within an in-house facility operated by trusted staff. But increasingly, data processing is moving out of the trusted zone and security mechanisms that used to be effective do not work any more. Secure computation holds the promise for a solution that will completely change the landscape of cyber security. With secure computation, calculations work on encrypted data directly: data can be kept encrypted all the time during the computation, so theft of data is harmless. Information only gets released under specific control of the holder of the encryption keys. Unfortunately, secure computation is not practical yet. Computing on encrypted data is often orders of magnitude slower than computing on plaintext data. Even with the latest technology, running the killer apps, which are often data-intensive, in secure computation is still a mission impossible.

In this talk, I will present two approaches that could lead to practical data-intensive secure computation. The first approach is by designing data structures. Traditionally, data structures have been widely used in computer science to improve performance of computation. However, in secure computation they have been largely overlooked in the past. I will show that data structures could be effective performance boosters in secure computation. Another approach is by using fully homomorphic encryption (FHE) technology. A common belief is that FHE is too inefficient to have any practical applications for the time being. Contrary to this common belief, I will show that in some cases FHE can actually lead to very efficient secure computation protocols.