JMU Artificial Intelligence and Machine Learning Seminar Series
Organized by: Hala Nelson and Nathan Sprague
There is hardly any area in our lives that has not been touched or transformed by Artificial Intelligence and Machine Learning.
This monthly seminar series showcases recent advances and open questions in these far reaching and rapidly advancing areas.
Topics include ethics and fairness, policy and regulation, technological advances, hardware, computing systems, methods and algorithms,
mathematical analysis, as well as Artificial Intelligence and Machine Learning applications in engineering, physics, astrophysics, chemistry,
materials science, biology, health sciences, earth and environmental sciences, transportation, operation systems,
social and behavioral sciences, and others. Undergraduate students are strongly encouraged to attend.
Third Thursday of each month: 4:10-5:00 pm Eastern Time (US and Canada)
Zoom meeting. Click here to register.
AI and ML Seminar Series Flyer
Please share our flyer: AI and ML Seminar Flyer
Spring 2021 Seminars
Jan 21, 2021 04:10 PM: Addressing Systemic Challenges in Peer Review
Talk by Dr. Nihar B. Shah from Carnegie Melon University. Nihar B. Shah is an Assistant Professor in the Machine Learning and Computer Science departments at Carnegie Mellon University (CMU).
His research interests include statistics, machine learning, information theory, and game theory, with a focus on applications to learning from people.
He is a recipient of an NSF CAREER Award 2020-25, the 2017 David J. Sakrison memorial prize from EECS Berkeley for a "truly outstanding and innovative PhD thesis",
the Microsoft Research PhD Fellowship 2014-16, the Berkeley Fellowship 2011-13,
the IEEE Data Storage Best Paper and Best Student Paper Awards for the years 2011/2012,
and the SVC Aiya Medal 2010, and has supervised the Best Student Paper at AAMAS 2019.
Abstract: Peer review is the backbone of scientific research, where research papers or grant proposals submitted by scientists are
evaluated by other experts in the field. Peer review also forms an integral part of employee evaluations in many organizations as well
as in grading in MOOCs and large classes. Peer review however faces a number of challenges which cause unfairness to the participants,
and degrades the overall quality of the process. This talk will present some systemic challenges in peer review and practical approaches to address them:
No prior background will be assumed.
- Errors: Errors in evaluation due to inappropriate assignment of reviewers to papers.
- Mis-calibration: Reviewers are lenient/strict/extreme/moderate etc.
- Dishonest behavior: In competitive settings, reviewers are incentivized to downgrade papers they are reviewing in
order to increase the relative performance of their own papers.
- Subjectivity: Different reviewers weigh various factors differently.
- Bias: The review is influenced by the identity of the authors.
Feb 17, 2021 03:10 PM: An unsupervised approach to characterizing users in online social media platforms
Talk by Dr. Dan Vilenchik from Ben-Gurion University.
Dan Vilenchik is an Assistant Professor in the School of Electrical and Computer Engineering in Ben-Gurion University.
His research interests include computational statistics and machine learning, with a focus on online social platforms and text analysis.
He received his PhD in Computer Science from Tel-Aviv University in 2008, winning the dean's award for excellent PhD dissertation.
Between 2008-2011 he was a postdoc at the EECS department at UC Berkeley and a Hedrick Assistant Professor at the math department at UCLA.
Abstract: Online social media channels play a central role in our lives. Characterizing users in social networks is a long-standing question,
dating back to the 50's when Katz and Lazarsfeld studied influence in "Mass Communication".
In the era of Machine Learning, this task is typically cast as a supervised learning problem,
where a target variable is to be predicted: age, gender, political incline, income, etc.
In this talk explore what can be achieved in an unsupervised manner.
Specifically, we harness Principal Component Analysis (PCA) to understand what underlying patterns and structures are inherent to some
social media platforms, but not to others, and why. We arrive at a Simpson-like paradox that may give us a deeper understanding of the
data-driven process of user characterization is such platforms.
Mar 18, 2021 04:10 PM: Graph Generative Adversarial Networks for High Energy Physics Data Generation
Talk by Ph.D. Candidate Raghav Kansal
from University of California, San Diego. Raghav Kansal is a PhD candidate in Physics at the University of California, San Diego (UCSD),
also working at the European Organisation for Nuclear Research (CERN). His research interests include particle physics and deep learning,
with a focus on graph-based networks and generative models for high energy particle collisions.
He graduated summa cum laude with a BS in Physics and Computer Engineering from UCSD,
and is a recipient of the 2019 John Holmes Malmberg Prize in Physics, and a 2019 IRIS-HEP Fellowship.
Abstract: Graph-based networks, with their ability to handle sparse, permutation invariant data with complex geometries,
have recently proven useful in a variety of disciplines. This includes high energy physics, where they have been successfully applied to important
classification and reconstruction tasks, however have yet to be explored for generation. We develop new graph-based generative models,
using the message passing neural network and generative adversarial network frameworks, for simulating datasets like those produced at the CERN Large Hadron
Collider (LHC). We demonstrate our model by training on and generating graphical representations of MNIST images, and jets of particles in proton-proton
collisions like those at the LHC.
Apr 22, 2021 05:10 PM: A PDE Interpretation of Prediction with Expert Advice
Talk by Dr. Nadejda Drenska from from University of Minnesota. Dr. Nadejda Drenska is a Minnesota Center for Financial and
Actuarial Mathematics postdoc at the School of Mathematics at the University of Minnesota. Her research interests include: nonlinear analysis, PDEs,
semi-supervised learning, repeated two-person games, graph theory, and applications in computer science.
Nadejda graduated with BS Applied Mathematics and BS in Mathematics, Honors, magna cum laude,
from Brown University and received the Rohn Truell prize for Outstanding Undergraduate Student. She obtained her PhD at the
Courant Institute of Mathematical Sciences at New York University and was awarded the Moses Green Award for Outstanding Interdisciplinary Studies from
the Courant Institute at NYU. Starting in July 2021, Dr. Nadejda will be joining John Hopkins University as Rufus Isaacs Postdoctoral Fellow.
Abstract: We study the problem of prediction of binary sequences with expert advice in the online setting,
which is a classic example of online machine learning. We interpret the binary sequence as the price history of a stock,
and view the predictor as an investor, which converts the problem into a stock prediction problem.
In this framework, an investor, who predicts the daily movements of a stock, and an adversarial market,
who controls the stock, play against each other over N turns.
The investor combines the predictions of n greater than two experts in order to make a decision about how much to invest at each turn,
and aims to minimize their regret with respect to the best-performing expert at the end of the game.
We consider the problem with history-depen-dent experts, in which each expert uses the previous d days of history of the market
in making their predictions. The prediction problem is played (in part) over a discrete graph called the d dimensional de Bruijn graph. We focus on
an appropriate continuum limit and using methods from optimal control,
graph theory, and partial differential equations, and we discuss strategies for the investor and the adversarial market.
We prove that the value function for this game, rescaled appropriately,
converges as N goes to infinity at a rate of O(N-1/2) (for C4 payoff functions) to the viscosity solution of a nonlinear degenerate parabolic PDE.
It can be understood as the Hamilton-Jacobi-Issacs equation for the two-person game.
As a result, we are able to deduce asymptotically optimal strategies for the investor.
This is joint work with Robert Kohn and Jeff Calder.
Time shown in Eastern Time (US and Canada)