Thursday, October 17, 2019
ISC 2017
Algorithm and Data Privacy in the Era of Big Data

Algorithm and Data Privacy in the Era of Big Data

Mahtab Mirmohseni (Assistant Prof.)

In dealing with big data, we often have to interact with some unreliable parties. These parties could be processors that are used to offload the computation on, or databases that store and maintain data. In both cases, these parties could be curious to gain some information about our data or even our algorithms. In this talk, we investigate these two scenarios.

1- Massive Matrix Computation with Unreliable Processors

We consider the problem of computing a function of some massive matrices using some distributed nodes with limited computing resources. In addition, these nodes are unreliable: they could be faulty, slow, and more importantly too curious i.e. some of them may collude to gain information about the data. One conventional approach to deal with these limitations is to concatenate job-splitting and secure computation. In this approach, the computation task is divided into some smaller sub-tasks, up to the capacity of each node, and then conventional approaches (i.e. secure multiparty computation) is used to execute each sub-task.
In this part, we argue that the concatenation of job-splitting and secure computation is inefficient. We propose a secure data sharing scheme, relying on new approaches of coded computing, that simultaneously, (i) reduces the size of data/computation to fit in each node, (2) secures the data against dishonest colluding nodes, (3) admits all basic operations such as addition and multiplication. This would allow us to securely compute any polynomial function of the massive matrices. The proposed scheme significantly improves concatenation solution in terms of number of nodes needed. The improvement can scale with the size of the problem.

2- Private Function Retrieval with Unreliable Databases

Motivated by the algorithm privacy concerns in a distributed computing system, in this part, we introduce the private function retrieval (PFR) problem, where a user wishes to retrieve a linear function of some messages that are assumed to be entirely available in some non-communicating replicated servers. However, these servers are curious to know about the function itself, which represents the user’s valuable algorithm. The objective is to minimize the communication load between the agent and the servers, where the function remains private. We propose an algorithm for PFR, exploiting the existence of several servers. The proposed algorithm is very efficient, and proven to optimum in some cases.

تاریخ به روزرسانی:
1397/05/27
تعداد بازدید:
97
Copyright @ 2010-2015 Iranian Society of Cryptology. All rights reserved.
Powered by DorsaPortal