This session was recorded in NYC on October 22nd, 2019 and can be viewed here: https://youtu.be/YCMFz955ajo
Reducing AI Bias Using Truncated Statistics
An emergent threat to the practical use of machine learning is the presence of bias in the data used to train models. Biased training data can result in models which make incorrect or disproportionately correct decisions, or that reinforce the injustices reflected in their training data. For example, recent works have shown that semantics derived automatically from text corpora contain human biases, and found that the accuracy of face and gender recognition systems are systematically lower for people of color and women. While the root causes of AI bias are difficult to pin down, a common cause of bias is the violation of the pervasive assumption that the data used to train models are unbiased samples of an underlying “test distribution,” which represents the conditions that the trained model will encounter in the future. Overcoming the bias introduced by the discrepancy between train and test distributions has been the focus of a long line of research in truncated Statistics. We provide computationally and statistically efficient algorithms for truncated density estimation and truncated linear, logistic and probit regression in high dimensions, through a general, practical framework based on Stochastic Gradient Descent. We illustrate the efficacy of our framework through several experiments.
Bio: David Eisenbud served as Director of MSRI from 1997 to 2007, and began a new term in 2013. He received his PhD in mathematics in 1970 at the University of Chicago under Saunders MacLane and Chris Robson, and was on the faculty at Brandeis University before coming to Berkeley, where he became Professor of Mathematics in 1997. He served from 2009 to 2011 as Director for Mathematics and the Physical Sciences at the Simons Foundation, and is currently on the Board of Directors of the Foundation. He has been a visiting professor at Harvard, Bonn, and Paris. Eisenbud’s mathematical interests range widely over commutative and non-commutative algebra, algebraic geometry, topology, and computer methods.
Eisenbud is Chair of the Editorial Board of the Algebra and Number Theory journal, which he helped found in 2006, and serves on the Board of the Journal of Software for Algebra and Geometry, as well as Springer-Verlag’s book series Algorithms and Computation in Mathematics.
Eisenbud was President of the American Mathematical Society from 2003 to 2005. He is a Director of Math for America, a foundation devoted to improving mathematics teaching. He has been a member of the Board of Mathematical Sciences and their Applications of the National Research Council, and is a member of the U.S. National Committee of the International Mathematical Union. In 2006, Eisenbud was elected a Fellow of the American Academy of Arts and Sciences.
2. 17 GAUSS WAY, BERKELEY, CALIFORNIA WWW.MSRI.ORG
Collaborative research at
the frontiers
3. What Distinguishes MSRI?
Exceptional Breadth of Scientific Activity
• MSRI programs span every aspect of fundamental
mathematics and many applications resulting from the
latest discoveries of advanced research
• MSRI is a center where new fields are born (MSRI
was among the first to recognize Random Matrix
Theory as a significant field), and where classical
topics such as symplectic geometry can have a major
revival.
• We connect mathematicians and math educators
through workshops on “Critical Issues in K-12
Education”
3
4. What Distinguishes MSRI?
A place where young talent flourishes
• MSRI is a place where young talent is mentored,
where careers are developed, and where
underrepresented groups are empowered
• Summer Graduate Schools include strong
contact between students and leaders
• Groundbreaking, sustained programs
empowering women and minorities (MSRI- UP,
SWiM, ADJOINT, Connections workshops...)
4
5. What Distinguishes MSRI?
MSRI is supported by the whole Math Community
• A network of 110 “Academic Sponsor” institutions – from the
US and abroad share in governance
• Impeccable advisors such as Dusa MacDuff, Fields Medalists
Charlie Fefferman, Andrei Okounkov, Terry Tao (Board
members and program organizers) and a rotating Scientific
Advisory Committee (SAC) help us reach into all parts of
fundamental mathematics
5
6. What Distinguishes MSRI?
MSRI is a pioneer in international engagement
• Increases connections between US researchers and the
international community
• In 2020, MSRI partnered with universities in Australia, Canada,
China, England, Greece, and Switzerland on Summer Graduate
Schools in emerging areas, using non-NSF funds and international
support.
• Co-founders of the Banff International Research Station (BIRS) and
Casa Matemática Oaxaca (CMO), where MSRI remains a full
partner in governance and in scientific interactions
6
7. What Distinguishes MSRI?
Public Understanding of Mathematics
• Congressional Briefings 2x/year
• Numberphile—most popular informal math channel on
YouTube: over 3M subscribers, 450M ”views”
(Numberphile.com)
• Films, such as Navajo Math Circles, and Secrets of the
Surface, the Mathematical Vision of Maryam Mirzakhani
• National Math Festival
• Mathical Books—a prize for children’s literature related to
mathematics
7
8. • To formalize and facilitate the flow of information between
the corporate world and MSRI we have developed a
corporate partnership program.
• Corporate members may choose from three tiers depending
on the level of interaction and sponsorship desired.
• Custom sponsorship and individualized packages are also
available.
8
Corporate Partners Program
10. Tier 1 $150,000/year
• MSRI-UP nonacademic employment opportunities information
session (lunch or dinner)
• Two at large nonacademic employment opportunities information
sessions at MSRI
• Invitation to the Academic Sponsors and MSRI Board of Trustees
evening (Spring)
• Invitations to Museion (distinguished donor) events
• Recognition on MSRI website
• Desk available for corporate research use at MSRI
10
Corporate Partners Program
11. Tier 2 $300,000/year
• All of the Tier 1 benefits plus:
– Named postdoc
– Principal program sponsor recognition and
participation opportunities for one semester-long
program
– Prominent sponsor recognition for all workshops
11
Corporate Partners Program
12. Tier 3 $500,000/year
• All of the Tier 2 benefits plus:
– Named research professor or 2 named postdocs
– Prominent sponsor recognition for MSRI summer
schools
12
Corporate Partners Program
13. Mathematical themes for current and future Programs
• Analysis and Mathematical Physics: Programs 1, 2, 3, 4,
7, 9
• Number Theory: Programs 5,6,10
• Dynamical Systems and Fluid Mechanics: Programs 7,8,9
• Probability: Programs 5,8,11
• Economics: Program 11
• Quantum Mechanics: Programs 1 and 3
• What Mathematics will be useful next??
13
Corporate Partners Program
14. For Reference: Current and future MSRI Programs:
1. Holomorphic Differentials in Mathematics and Physics (Fall 2019)
2. Microlocal Analysis (Fall 2019)
3. Quantum Symmetries (Spring 2020)
4. Higher Categories and Categorification (Spring 2020)
5. Random and Arithmetic Structures in Topology (Fall 2020)
6. Decidability, Definability and Computability in Number Theory (Fall 2020)
7. Mathematical Problems in Fluid Dynamics (Spring 2021)
8. Universality and Integrability in Random Matrix Theory and Interacting
Particle Systems (Fall 2021)
9. Complex Dynamics: from Special Families to Natural Generalizations in one
and Several Variables (Spring 2022)
10. Algebraic Cycles, L-Values, and Euler Systems (Spring 2023)
11. Mathematics and Computer Science of Market and Mechanism Design (Fall
2023)
14
Corporate Partners Program
15. 17 GAUSS WAY, BERKELEY, CALIFORNIA WWW.MSRI.ORG
From MSRI…
The view is global!
Notas del editor
Please feel free to interrupt with questions as I speak!
A little history:
MSRI was born in an NSF competition, and had its first programs in 1982, 37 years ago.
Over the last 15 years it has become a true public-private partnership
Rather than the basics, I want to start with some of the things that I think make MSRI special.
*We strive to be the premier Collaborative Math Institute concentrating on fundamental math, not ignoring applications
Boldness in choosing fields such as Random Matrix Theory
Range of workshops from Number Theory to Insect Navigation
Program examples:
Algebraic Geometry, The Langlands Program, plasma physics. Fluid Dynamics.
Summer Graduate Schools examples:
Data Science (2019), Machine Learning (2020), Animal Behavior at Janelia Laboratories of HHMI;
Hot topics workshops examples:
Optimal transport and applications; Machine learning and statistics, (2020); Workshop on Insect Navigation at HHMI
We take mentoring VERY seriously.
Highly structured mentoring of Postdoc; we work hard to connect Postdocs to distinguished participants
Rapidly expanding system of summer graduate schools
Groundbreaking programs for women and minorities; often later imitated (HRAC) or Shared (Blackwell-Tapia) with other institutes
Many stories of career-changing experience; some in postdoc slide later
Wide Community engagement and support:
10 Academic sponsors in 1982 110 today. Shared governance.
Wide range of institution types are welcome, elite places like Harvard and MIT, but also some without PhD programs such as SF State and Portland State.
Some of the worlds most distiinguished mathematicians feel it worth their time to participate in our governance structures
To remain world leaders in math, the U.S. *must* cultivate international contacts.
International visitors to MSRI greatly add to the programs, form connections with US mathematicians
Our Summer Graduates schools abroad (jointly funded with the host countries) create bonds between US graduate students and some of the brightest students abroad
MSRI played an important role in founding the Banff International Research Station in Canada, and the Casa Matematica Oaxaca in Mexico, funded jointly by NSF and Canada, Mexico: ongoing collaborations enrich MSRI as well as these great resources for the math community
To remain world leaders in math, the U.S. *must* cultivate international contacts.
International visitors to MSRI greatly add to the programs, form connections with US mathematicians
Our Summer Graduates schools abroad (jointly funded with the host countries) create bonds between US graduate students and some of the brightest students abroad
MSRI played an important role in founding the Banff International Research Station in Canada, and the Casa Matematica Oaxaca in Mexico, funded jointly by NSF and Canada, Mexico: ongoing collaborations enrich MSRI as well as these great resources for the math community
You’ll see that we do many things; but our largest efforts and expenditures, by far, are on science!
Do we represent the NSF and the mathematical community well? An important question.