Campus Event Calendar
Event Entry
What and Who
Algorithms, Complexity, and Bias
Nisheeth Vishnoi
EPF Lausanne
MPI Colloquium Series Distinguished Speaker
Nisheeth Vishnoi is a faculty member of the School of Computer and
Communication Sciences at École Polytechnique Fédérale de Lausanne. He
is also an associate of the International Center for Theoretical
Sciences, Bangalore, an adjunct faculty member of IIT Delhi and IIT
Kanpur, and a co-founder of the Computation, Nature, and Society
ThinkTank. His research focuses on foundational problems in
algorithms, optimization, statistics, and complexity, and how tools
from these areas can be used to address emerging algorithmic questions
in society, nature, and machine learning. Topics from these areas that
he is currently interested in include algorithmic fairness,
understanding emergent behavior in biological systems, and developing
algorithms that can go beyond the worst-case in machine learning. He
is the recipient of the Best Paper Award at FOCS 2005, the IBM
Research Pat Goldberg Memorial Award for 2006, the Indian National
Science Academy Young Scientist Award for 2011 and the IIT Bombay
Young Alumni Achievers Award for 2016.
AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI
Public Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Thursday, 7 December 2017
14:15
75 Minutes
E1 4
024
Saarbrücken
Abstract
Bias is an increasingly observed phenomenon in the world of machine
learning (ML) and artificial intelligence (AI): From gender bias in
image search, racial bias in court bail pleas, to biases in
worldviews depicted in personalized newsfeeds. At the core, what is
powering today’s AI/ML are algorithms for fundamental computational
problems such as classification, data summarization, ranking, and
online learning. Such algorithms have traditionally been designed with
the goal of maximizing some notion of “utility” and identifying or
controlling bias in their output has not been a consideration. As a
consequence, they have the ability to pass (or even worsen) the bias
from the input to the output.
In this talk, I will explain the emergence of bias in algorithmic
decision making and address some of the challenges towards developing
a systematic algorithmic framework to control biases and the
corresponding complexity barriers in the aforementioned problems. As a
concrete example, I will focus on the problem of data summarization
and show how the pursuit of “unbiased'' algorithms for this problem
has led to connections between entropy, continuous optimization,
polynomials, and the complexity of the Kadison-Singer problem.
learning (ML) and artificial intelligence (AI): From gender bias in
image search, racial bias in court bail pleas, to biases in
worldviews depicted in personalized newsfeeds. At the core, what is
powering today’s AI/ML are algorithms for fundamental computational
problems such as classification, data summarization, ranking, and
online learning. Such algorithms have traditionally been designed with
the goal of maximizing some notion of “utility” and identifying or
controlling bias in their output has not been a consideration. As a
consequence, they have the ability to pass (or even worsen) the bias
from the input to the output.
In this talk, I will explain the emergence of bias in algorithmic
decision making and address some of the challenges towards developing
a systematic algorithmic framework to control biases and the
corresponding complexity barriers in the aforementioned problems. As a
concrete example, I will focus on the problem of data summarization
and show how the pursuit of “unbiased'' algorithms for this problem
has led to connections between entropy, continuous optimization,
polynomials, and the complexity of the Kadison-Singer problem.
Contact
Petra Schaaf
5000
--email hidden
passcode not visible
logged in users only
Petra Schaaf, 11/29/2017 12:27 -- Created document.