Title: Algorithmic decision making and the cost of fairness
Abstract: Algorithms are now regularly used to decide whether defendants awaiting trial are too dangerous to be released back into the community. In some cases, black defendants are substantially more likely than white defendants to be incorrectly classified as high risk. To mitigate such disparities, several techniques have recently been proposed to achieve algorithmic fairness. We reformulate algorithmic fairness as constrained optimization: the objective is to maximize public safety while satisfying formal fairness constraints designed to reduce racial disparities. We show that for several past definitions of fairness, the optimal algorithms that result require detaining defendants above race-specific risk thresholds. We further show that the optimal unconstrained algorithm requires applying a single, uniform threshold to all defendants. The unconstrained algorithm thus maximizes public safety while also satisfying one important understanding of equality: that all individuals are held to the same standard, irrespective of race. Because the optimal constrained and unconstrained algorithms generally differ, there is tension between improving public safety and satisfying prevailing notions of algorithmic fairness. By examining data from Broward County, Florida, we show that this trade-off can be large in practice. We focus on algorithms for pretrial release decisions, but the principles we discuss apply to other domains, and also to human decision makers carrying out structured decision rules.
Paper:
https://5harad.com/papers/fairness.pdfBio: Sharad Goel is an assistant professor at Stanford University in the School of Engineering. His primary area of research is computational social science, an emerging discipline at the intersection of computer science, statistics, and the social sciences. Sharad is particularly interested in applying modern computational and statistical techniques to understand and improve public policy. Topics he's recently worked on include stop-and-frisk, tests for racial bias, algorithmic fairness, swing voting, voter fraud, filter bubbles, and online privacy.
--------------------------------------------------
Friday 12pmRaghavan Komondoor (Indian Institute of Science)
Title: Testing and Analysis of Web Applications using Page Models
Abstract: Web applications are difficult to analyze using code-based tools because
data-flow and control-flow through the application occurs via both
server-side code and client-side pages. Client-side pages are typically
specified in a scripting language that is different from the main
server-side language; moreover, the pages are generated dynamically from
the scripts. To address these issues we propose a static-analysis approach
that automatically constructs a ``model'' of each page in a given
application. A page model is a code fragment in the same language as the
server-side code, which faithfully over-approximates the possible elements
of the page as well as the control-flows and data-flows due to these
elements. The server-side code in conjunction with the page models then
becomes a standard (non-web) program, thus amenable to analysis using
standard code-based tools.
We have implemented our approach in the context of J2EE applications. We
demonstrate the versatility and usefulness of our approach by applying
three standard analysis tools on the resultant programs from our approach:
a concolic-execution based model checker (JPF), a dynamic fault
localization tool (Zoltar), and a static slicer (Wala).
Speaker bio:
Raghavan Komondoor is an Associate Professor at the Department of Computer
Science and Automation, Indian Institute of Science (IISc), Bangalore. He
obtained his PhD in Computer Sciences from University of Wisconsin-Madison,
advised by Prof. Susan Horwitz. He worked for IBM Research for
several years before moving to IISc. His areas of interest are program
analysis, programming tools, and formal methods for software engineering.
---------------------------
Monday 12pmSamuel Drews - CAV PRACTICE TALK
Title: Repairing Decision-Making Programs under Uncertainty
---------------------------
Tuesday 12pmBorzoo Bonakdarpour (McMaster University)
Title: Hypermonitirng Hyperproperties
Abstract: Cybersecurity is an area of information technology where dependability plays a crucial role. This is because even a short transient violation of security policies may result in leaking private or highly sensitive information, compromise safety, or lead to the interruption of vital public or social services. This talk will go over a general runtime monitoring technique for a rich class of security polices (e.g., information flow, etc) that cannot be expressed by traditional trace-based specification languages. To this end, we employ the logic HyperLTL derived from Clarkson and Schneider's theory of hyperproperties. We first define the notion of monitorability and identify monitorable policies. Then, I will describe two monitoring techniques, called hypermonitoring.
Bio: Borzoo Bonakdarpour is currently an assistant professor at the Department of Computing and Software at McMaster University, Canada. His research interests include runtime monitoring of distributed systems and security/privacy policies, power-aware algorithms, and program synthesis. His work in these areas have received multiple best paper awards and nominations from highly prestigious conferences. His tools SYCRAFT and ASSESS are capable of synthesizing fault-tolerant and self-stabilizing distributed protocols. He chaired the Technical Program Committee of the SSSâ16 and RVâ14 conferences. His Ph.D. dissertation, "Automated Revision of Distributed and Real-Time Programs", was nominated for the 2010 ACM Doctoral Dissertation Award.
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}