\usepackage{beamerthemeplaincu}

%\usepackage[T1]{fontenc}

%\usepackage[latin1]{inputenc}

\renewcommand{\slidecaption}{APP 08, King's College London, 26 November 2013}

  Office: & S1.27 (1st floor Strand Building)\\

\frametitle{Man-in-the-Middle}

\frametitle{Facebook Privacy}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[t]

\frametitle{Privacy}

\begin{minipage}{1.05\textwidth}

\begin{itemize}

\item we \alert{do} want that government data is made public (free maps for example)

\item we \alert{do not} want that medical data becomes public (similarly tax data, school 

records, job offers)\bigskip

\item personal information can potentially lead to fraud 

(identity theft)

\end{itemize}\pause

{\bf ``The reality'':}

\only<2>{\begin{itemize}

\item London Health Programmes lost in June last year unencrypted details of more than 8 million people

(no names, but postcodes and details such as gender, age and ethnic origin)

\end{itemize}}

\only<3>{\begin{itemize}

\item also in June last year, Sony got hacked: over 1M users' personal information, including passwords, email addresses, home addresses, dates of birth, and all Sony opt-in data associated with their accounts.

\end{itemize}}

\end{minipage}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Privacy and Big Data}

Selected sources of ``Big Data'':\smallskip{}

\begin{itemize}

\item Facebook 

\begin{itemize}

\item 40+ Billion photos (100 PB)

\item 6 Billion messages daily (5 - 10 TB)

\item 900 Million users  

\end{itemize}

\item Common Crawl

\begin{itemize}

\item covers 3.8 Billion webpages (2012 dataset)

\item 50 TB of data

\end{itemize}

\item Google

\begin{itemize}

\item 20 PB daily (2008)

\end{itemize}

\item Twitter

\begin{itemize}

\item 7 Million users in the UK

\item a company called Datasift is allowed to mine all tweets since 2010

\item they charge 10k per month for other companies to target advertisement

\end{itemize}

\end{itemize}\pause

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Cookies\ldots}

``We have published a new cookie policy. It explains what cookies are 

and how we use them on our site. To learn more about cookies and 

their benefits, please view our cookie policy.\medskip

If you'd like to disable cookies on this device, please view our information 

pages on 'How to manage cookies'. Please be aware that parts of the 

site will not function correctly if you disable cookies. \medskip

By closing this 

message, you consent to our use of cookies on this device in accordance 

with our cookie policy unless you have disabled them.''

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Scare Tactics}

The actual policy reads:\bigskip

``As we explain in our Cookie Policy, cookies help you to get the most 

out of our websites.\medskip

If you do disable our cookies you may find that certain sections of our 

website do not work. For example, you may have difficulties logging in 

or viewing articles.''

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Netflix Prize}

Anonymity is \alert{necessary} for privacy, but \alert{not} enough!\bigskip

\begin{itemize}

\item Netflix offered in 2006 (and every year until 2010) a 1 Mio \$ prize for improving their movie rating algorithm

\item dataset contained 10\% of all Netflix users (appr.~500K)

\item names were removed, but included numerical ratings as well as times of rating

\item some information was \alert{perturbed} (i.e., slightly modified)

\end{itemize}

\hfill{\bf\alert{All OK?}}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Re-identification Attack}

Two researchers analysed the data: 

\begin{itemize}

\item with 8 ratings (2 of them can be wrong) and corresponding dates that can have a margin 14-day error, 98\% of the

records can be identified

\item for 68\% only two ratings and dates are sufficient (for movie ratings outside the top 500)\bigskip\pause

\item they took 50 samples from IMDb (where people can reveal their identity)

\item 2 of them uniquely identified entries in the Netflix database (either by movie rating or by dates)

\end{itemize}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{}

\begin{itemize}

\item Birth data, postcode and gender (unique for\\ 87\% of the US population)

\item Preferences in movies (99\% of 500K for 8 ratings)

\end{itemize}\bigskip

Therefore best practices / or even law (HIPAA, EU): 

\begin{itemize}

\item only year dates (age group for 90 years or over), 

\item no postcodes (sector data is OK, similarly in the US)\\

\textcolor{gray}{no names, addresses, account numbers, licence plates}

\item disclosure information needs to be retained for 5 years

\end{itemize}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}<2>[c]

\frametitle{How to Safely Disclose Information?}

\only<1>{

\begin{itemize}

\item Assume you make a survey of 100 randomly chosen people.

\item Say 99\% of the surveyed people in the 10 - 40 age group have seen the

Gangnam video on youtube.\bigskip

\item What can you infer about the rest of the population? 

\end{itemize}}

\only<2>{

\begin{itemize}

\item Is it possible to re-identify data later, if more data is released. \bigskip\bigskip\pause

\item Not even releasing only  aggregate information prevents re-identification attacks.

(GWAS was a public database of gene-frequency studies linked to diseases;

you only needed partial DNA information  in order

to identify whether an individual was part of the study --- DB closed in 2008) 

\end{itemize}}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Differential Privacy}

\begin{center}

User\;\;\;\;    

\begin{tabular}{c}

tell me \bl{$f(x)$} $\Rightarrow$\\

$\Leftarrow$ \bl{$f(x) + \text{noise}$}

\end{tabular}

\;\;\;\;\begin{tabular}{@{}c}

Database\\

\bl{$x_1, \ldots, x_n$}

\end{tabular}

\end{center}

\begin{itemize}

\item \bl{$f(x)$} can be released, if \bl{$f$} is insensitive to

individual entries  \bl{$x_1, \ldots, x_n$}\\

\item Intuition: whatever is learned from the dataset would be learned regardless of whether

\bl{$x_i$} participates\bigskip\pause 

\item Noised needed in order to prevent queries:\\ Christian's salary $=$ 

\begin{center}

\bl{\large$\Sigma$} all staff $-$  \bl{\large$\Sigma$} all staff $\backslash$ Christian

\end{center} 

\end{itemize}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[c]

\frametitle{Adding Noise}

Adding noise is not as trivial as one would wish:

\begin{itemize}

\item If I ask how many of three have seen the Gangnam video and get a result

as follows 

\begin{center}

\begin{tabular}{l|c}

Alice & yes\\

Bob & no\\

Charlie & yes\\

\end{tabular}

\end{center}

then I have to add a noise of \bl{$1$}. So answers would be in the

range of \bl{$1$} to \bl{$3$}

\bigskip

\item But if I ask five questions for all the dataset (has seen Gangnam video, is male, below 30, \ldots),

then one individual can change the dataset by \bl{$5$}

\end{itemize}

\end{frame}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\mode<presentation>{

\begin{frame}[t]

\frametitle{\begin{tabular}{@{}c@{}}Tor\end{tabular}}

\begin{itemize}

\item initially developed by US Navy Labs, but then opened up to the world 

\item network of proxy nodes

\item a Tor client establishes a ``random'' path to the destination server (you cannot trace back where the information came from)\bigskip\pause

\end{itemize}

\only<2>{

\begin{itemize}