Skip to content

Commit

Permalink
draft for lecture 3
Browse files Browse the repository at this point in the history
  • Loading branch information
@lisahdd007
lisahdd007 committed Jun 4, 2020
1 parent b7cb8bc commit 7954ce1
Show file tree
Hide file tree
Showing 6 changed files with 138 additions and 60 deletions.
Binary file added img/rml_diagram_lec3_hilite.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added img/rml_diagram_lec4_hilite.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file modified tex/lecture_3.pdf
View file
Binary file not shown.
121 changes: 80 additions & 41 deletions tex/lecture_3.tex
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,4 @@
\documentclass[11pt,
aspectratio=169,
hyperref={colorlinks}
]{beamer}
\documentclass[11pt,aspectratio=169,hyperref={colorlinks}]{beamer}
\usetheme{Singapore}
\usecolortheme[snowy, cautious]{owl}

Expand All @@ -12,12 +9,16 @@
\usepackage{hyperref}
\hypersetup{
colorlinks=true,
urlcolor=[rgb]{1,0,1},
linkcolor=[rgb]{1,0,1}}
urlcolor=[rgb]{0,0,0.61},
linkcolor=[rgb]{0,0,0.61}}
\usepackage[natbib=true,style=numeric,backend=bibtex,useprefix=true]{biblatex}
\usepackage{blindtext}

\definecolor{OwlGreen}{RGB}{75,0,130}
%\usepackage{blindtext}

%----------------------------------------------------------------------------------
\definecolor{OwlGreen}{RGB}{51,0,102}
%----------------------------------------------------------------------------------

\setbeamertemplate{bibliography item}{}
\setbeamerfont{caption}{size=\footnotesize}
\setbeamertemplate{frametitle continuation}{}
Expand All @@ -32,13 +33,14 @@

\renewcommand*{\thefootnote}{\fnsymbol{footnote}}

\author{\copyright\hspace{1pt}Patrick Hall}
\title{Discrimination in Machine Learning\footnote{\tiny{This material is shared under a \href{https://creativecommons.org/licenses/by/4.0/deed.ast}{CC By 4.0 license} which allows for editing and redistribution, even for commercial purposes. However, any derivative work should attribute the author and H2O.ai.}}\footnote{\tiny{This presentation is not, and should not be construed as, legal advice or requirements for regulatory compliance.}}}
\subtitle{\scriptsize{}}
%\logo{\includegraphics[height=8pt]{img/h2o_logo.png}}
\institute{\href{https://www.h2o.ai}{H\textsubscript{2}O.ai}}
%------------------------------------------------------------------------------------------

\author{Patrick Hall}
\title{Responsible Machine Learning\footnote{\tiny{This material is shared under a \href{https://creativecommons.org/licenses/by/4.0/deed.ast}{CC By 4.0 license} which allows for editing and redistribution, even for commercial purposes. However, any derivative work should attribute the author.}}}
\subtitle{Lecture 3: Fairness}
\institute{The George Washington University}
\date{\today}
\subject{Subject}


\begin{document}

Expand All @@ -52,27 +54,45 @@

\end{frame}


%-------------------------------------------------------------------------------
\section{Why?}
\section{Introduction}
%-------------------------------------------------------------------------------

\subsection*{}

\begin{frame}

\frametitle{A Responsible Machine Learning Workflow\footnote{\href{https://www.mdpi.com/2078-2489/11/3/137/htm}{\textit{A Responsible Machine Learning Workflow}}}}

\begin{figure}[htb]
\begin{center}
\includegraphics[height=150pt]{../img/rml_diagram_lec3_hilite.png}
\label{fig:blueprint}
\end{center}
\end{figure}

\end{frame}


\begin{frame}

\frametitle{Why Care About Discrimination in ML?}

\begin{itemize}
\Large
\item \textbf{Reputational risk}: Upon encountering a perceived unethical ML system, 34\% of consumers are likely to, ``stop interacting with the company.''\footnote{\scriptsize{See: \href{https://www.capgemini.com/research/why-addressing-ethical-questions-in-ai-will-benefit-organizations/}{Why addressing ethical questions in AI will benefit organizations}.}}
\item \textbf{Reputational risk}:
\begin{itemize}
\Large
\item{\normalsize{Upon encountering a perceived unethical ML system, 34\% of consumers are likely to, ``stop interacting with the company.''\footnote{\scriptsize{See: \href{https://www.capgemini.com/research/why-addressing-ethical-questions-in-ai-will-benefit-organizations/}{Why addressing ethical questions in AI will benefit organizations}.}}}}
\end{itemize}
\item Non-compliance fines and litigation costs.
\item Responsible practice of ML.
\end{itemize}

\end{frame}

\subsection*{}



\begin{frame}

\frametitle{Elements of Responsible Practice of ML}
Expand All @@ -85,9 +105,8 @@

\end{frame}


%-------------------------------------------------------------------------------
\section{What?}
\section{Discrimination \& Bias}
%-------------------------------------------------------------------------------

\subsection*{}
Expand All @@ -99,27 +118,35 @@
\begin{itemize}
\Large
\item Almost all data, statistical models, and machine learning (ML) models encode different types of \textit{bias}, i.e., systematic misrepresentations of reality.\\
\item Sometimes, bias is helpful, e.g. shrunken, robust $\beta_j$ coefficients in penalized linear models.\\
\item Sometimes, bias is helpful.
\begin{itemize}
\Large
\item{shrunken and robust $\beta_j$ coefficients in penalized linear models} \end{itemize}
\item Other types of bias might be unwanted, unhelpful, or illegal discrimination.
\end{itemize}

\end{frame}

\subsection*{}

\begin{frame}
\begin{frame}[allowframebreaks]{Title}

\frametitle{What is Discrimination in ML?}

\scriptsize
\noindent In some applications\footnote{\tiny{e.g., Under the Equal Credit Opportunity Act (ECOA), as implemented by Regulation B, and the Fair Credit Reporting Act (FCRA})}, model predictions should \textit{ideally} be independent of demographic group membership.\\

\noindent In some applications\footnote{\small{e.g., Under the Equal Credit Opportunity Act (ECOA), as implemented by Regulation B, and the Fair Credit Reporting Act (FCRA})}, model predictions should \textbf{\textit{ideally}} be independent of demographic group membership.\\
\vspace{5pt}
\noindent In these applications, a model exhibits discrimination if:
\begin{enumerate}
\item Demographic group membership is not independent of the likelihood of receiving a favorable or accurate model prediction.
\item Membership in a \textit{subset} of a demographic group is not independent of the likelihood of receiving a favorable or accurate model prediction (i.e., \textit{local bias}).\cite{hall2019guidelines}
\end{enumerate}
\noindent Several forms of discrimination may manifest in ML, including:
\end{frame}

\begin{frame}

\frametitle{What is Discrimination in ML?}

\noindent \Large Several forms of discrimination may manifest in ML, including:
\begin{itemize}
\item Overt discrimination, i.e. \textit{disparate treatment}.
\item Unintentional discrimination, i.e. \textit{disparate impact} (DI).
Expand Down Expand Up @@ -162,51 +189,63 @@

\begin{frame}

\frametitle{What is Discrimination in ML?}
\frametitle{Common Metrics of Discrimination in ML}

Many kinds of group disparities can be measured, e.g.:\\
Common metrics of \textbf{\textit{group}} disparities:\\
\begin{itemize}
\item Accuracy disparity: $\frac{\text{accuracy}_p}{\text{accuracy}_r}$
\item Adverse impact ratio: $\frac{\text{\% accepted}_p }{ \text{\% accepted}_r}$
\item Adverse impact ratio: $\frac{\text{\% accepted}_p }{ \text{\% accepted}_r}$
\item Marginal effect: $\text{\% accepted}_p - \text{\% accepted}_r$
\item Standardized mean difference: $\frac{\bar{\hat{y}}_p - \bar{\hat{y}}_r}{\sigma_{\hat{y}}}$
\end{itemize}
\noindent where, $p \equiv \text{protected group}$ and $r \equiv \text{reference group}$ (often white males),\\
\noindent
\scriptsize where, $p \equiv \text{protected group}$ and $r \equiv \text{reference group}$ (often white males),\\
\vspace{5pt}
$\text{\% accepted}_\text{group} = 100 \cdot \frac{\text{tn}_\text{group}~+~\text{fn}_\text{group}}{N_\text{group}}$, and $\text{accuracy}_\text{group} = \frac{\text{tp}_\text{group}~+~\text{tn}_\text{group}}{N_\text{group}}$.\textsuperscript{$\mathparagraph$}\\
\vspace{10pt}
Local bias is much trickier to measure ... and often an unmitigated risk for consumer-facing ML systems.
\normalsize
\textbf{\textit{Local bias}} is much trickier to measure and often an unmitigated risk for consumer-facing ML systems.
\end{frame}

%-------------------------------------------------------------------------------
\section{How?}
\section{Remediation}
%-------------------------------------------------------------------------------

\subsection*{}

\begin{frame}

\frametitle{How to Fix Discrimination in ML?}
\scriptsize
\noindent \textbf{Fix the process}: ensure diversity of experience in design, training, and review of ML systems.\\
\noindent \textbf{Fix the data}:
\begin{itemize}\scriptsize
\Large \noindent
\textbf{Fix the process}: ensure diversity of experience in design, training, and review of ML systems.\\
\noindent
\textbf{Fix the data}:
\begin{itemize}
\item Collect demographically representative training data.
\item Select features judiciously, e.g. using \texttt{time\_on\_file} as an input variable as opposed to \texttt{bankruptcy\_flag}.\textsuperscript{$\mathparagraph$}
\item Sample and reweigh training data to minimize discrimination.\cite{kamiran2012data}
\end{itemize}
\noindent \textbf{Fix the model}:
\begin{itemize}\scriptsize

\end{frame}


\begin{frame}

\frametitle{How to Fix Discrimination in ML?}

\noindent
\textbf{Fix the model}:
\begin{itemize}
\item Consider fairness metrics when selecting hyperparameters and cutoff thresholds.
\item Train fair models directly:
\begin{itemize}\scriptsize
\begin{itemize}
\item Learning fair representations (LFR) and adversarial de-biasing.\cite{zemel2013learning}, \cite{zhang2018mitigating}
\item Use dual objective functions that consider both accuracy and fairness metrics.
\end{itemize}
\item Edit model mechanisms to ensure less biased predictions, e.g. with \href{https://github.com/interpretml/interpret}{GA2M} models.
\end{itemize}
\noindent \textbf{Fix the predictions}:
\begin{itemize}\scriptsize
\begin{itemize}
\item Balance model predictions, e.g. reject-option classification.\cite{kamiran2012decision}
\item Correct or override predictions with model assertions or appeal mechansims.\cite{hall2019guidelines}, \cite{kangdebugging}
\end{itemize}
Expand Down
Binary file modified tex/lecture_4.pdf
View file
Binary file not shown.
77 changes: 58 additions & 19 deletions tex/lecture_4.tex
View file
Original file line number Diff line number Diff line change
Expand Up @@ -14,19 +14,20 @@
\usepackage{hyperref}
\hypersetup{
colorlinks=true,
urlcolor=[rgb]{1,0,1},
linkcolor=[rgb]{1,0,1}}
urlcolor=[rgb]{0,0,0.61},
linkcolor=[rgb]{0,0,0.61}}

\usepackage[natbib=true,style=numeric,backend=bibtex,useprefix=true]{biblatex}
%------------------------------------------------------------------------------

\usepackage[natbib=true,style=numeric,backend=bibtex,useprefix=true]{biblatex}
%\setbeamercolor*{bibliography entry title}{fg=black}
%\setbeamercolor*{bibliography entry location}{fg=black}
%\setbeamercolor*{bibliography entry note}{fg=black}
\definecolor{OwlGreen}{RGB}{75,0,130} % easier to see
\setbeamertemplate{bibliography item}{\insertbiblabel}
\setbeamerfont{caption}{size=\footnotesize}
\setbeamertemplate{frametitle continuation}{}

\definecolor{OwlGreen}{RGB}{51,0,102} % easier to see
\setcounter{tocdepth}{1}
\renewcommand*{\bibfont}{\scriptsize}
\addbibresource{lecture_4.bib}
Expand All @@ -37,12 +38,14 @@
\setbeamertemplate{footline}{%
\raisebox{5pt}{\makebox{\hfill\makebox[20pt]{\color{gray}
\scriptsize\insertframenumber}}}\hspace*{5pt}}

\author{\copyright\hspace{1pt}Patrick Hall\footnote{\tiny{This material is shared under a \href{https://creativecommons.org/licenses/by/4.0/deed.ast}{CC By 4.0 license} which allows for editing and redistribution, even for commercial purposes. However, any derivative work should attribute the author and H2O.ai.}}}
\title{Machine Learning Models as an Attack Surface}
\institute{\href{https://www.h2o.ai}{H\textsubscript{2}O.ai}}


\author{Patrick Hall}
\title{Responsible Machine Learning\footnote{\tiny{This material is shared under a \href{https://creativecommons.org/licenses/by/4.0/deed.ast}{CC By 4.0 license} which allows for editing and redistribution, even for commercial purposes. However, any derivative work should attribute the author.}}}
\subtitle{Lecture 4: Security}
\institute{The George Washington University}
\date{\today}
\subject{Security of Machine Learning Models}


\begin{document}

Expand All @@ -55,22 +58,58 @@
\tableofcontents{}

\end{frame}



%-------------------------------------------------------------------------------
\section{Why?}
\section{Overview}
%-------------------------------------------------------------------------------
\subsection{Workflow} %just for progress indicator

\begin{frame}

\frametitle{A Responsible Machine Learning Workflow\footnote{\href{https://www.mdpi.com/2078-2489/11/3/137/htm}{\textit{A Responsible Machine Learning Workflow}}}}

\begin{figure}[htb]
\begin{center}
\includegraphics[height=150pt]{../img/rml_diagram_lec4_hilite.png}
\label{fig:blueprint}
\end{center}
\end{figure}

\end{frame}

\begin{frame}

\subsection{Why Attack}

\begin{frame}

\frametitle{Why Attack Machine Learning Models?}
\frametitle{Why Attack Machine Learning Models?}
Hackers, malicious or extorted insiders, and their criminal associates or organized extortionists, seek to:
\begin{itemize}
\item induce beneficial outcomes from a predictive or pattern recognition model or induce negative outcomes for others. %(loans, insurance policies, jobs, favorable criminal risk assessments, or others)
\item commit corporate espionage.
\item steal intellectual property including models and data.
\end{itemize}
\end{frame}
\begin{itemize}
\item induce beneficial outcomes from a predictive or pattern recognition model or induce negative outcomes for others. %(loans, insurance policies, jobs, favorable criminal risk assessments, or others)
\item commit corporate espionage.
\item steal intellectual property including models and data.
\end{itemize}
\end{frame}


\subsection{Types of Attack} %just for progress indicator

\begin{frame}

\frametitle{Types of Security Risks and Attacks}

\begin{itemize}
\Large
\item{Data Poisoning}
\item{Backdoors and Watermarks}
\item{Surrogate Model Inversion}
\item{Membership Inference}
\item{Adversarial Example}
\item{Impersonation Attacks}
\end{itemize}

\end{frame}
%-------------------------------------------------------------------------------
\section{Attacks}
%-------------------------------------------------------------------------------
Expand Down

0 comments on commit 7954ce1

Please sign in to comment.