A survey for differentially private learning

PhD Qualifying Examination


Title: "A survey for differentially private learning"

by

Mr. Peng PENG


Abstract:

Machine learning can be used to learn behaviors from empirical data, which 
is instrumental for building more sophisticated intelligent systems in 
many other disciplines such as data mining, bioinformatics and human 
cognition. However, we may suffer from private information disclosure when 
the learning results are released to the public. In the famous AOL event, 
a woman's identity was exposed after hundreds of the searching records she 
conducted over a three-month period are analyzed by data analysts. 
Motivated by the issue of privacy breaches, some researchers propose a new 
framework, which combines the learning process and privacy preserving 
computation. Specifically, in this survey, we focus on differentially 
private learning, including the learning algorithms that satisfy the 
notion of differential privacy, which is regarded as the gold standard in 
the privacy preserving community.

The fundamental objective of differentially private learning is to 
tradeoff between privacy and utility. That is, protect the privacy of 
individuals whose information can be found in the data set, while 
performing a suboptimal solution to the original non-private optimal 
solution.

In the following, we start by introducing the primitive definition of 
private learning, and describing a high level description of the 
combination between learning and differential privacy. Then, we show 
theoretical results in private learning and compare them with their 
non-private equivalents in order to give intuition on how differential 
privacy affects the utility in different settings. Moreover, we enumerate 
multiple specific private learning algorithms, which provide computational 
efficient implements for real-world applications. Finally, we end this 
survey by the conclusion and future works.


Date:                   Tuesday, 20 December 2011

Time:                   2:00pm - 4:00pm

Venue:                  Room 3501
                         lifts 25/26

Committee Members:	Dr. Raymong Wong (Supervisor)
                         Prof. James Kwok (Chairperson)
 			Dr. Wilfred Ng
 			Prof. Dit-Yan Yeung


**** ALL are Welcome ****