Abstract: Very few individuals, like Delbert Ray Fulkerson, have defined a field. Ray was a mathematician who pioneered research in theory and applications. This presentation will summarize for a general audience some of his fundamental contributions to flows on networks, linear and integer programming, and combinatorial analysis. With Lester Ford, he developed network flow theory, and with George Dantzig and others what  he developed methods for solving linear and integer programming, including such topics as primal-dual methods, cutting planes, and column generation for solving large scale optimization problems. He made signature contributions to extreme combinatorics, with novel and elegant results on pairs of polyhedra and perfect graphs. He contributed to applications in project scheduling, tanker scheduling, railway planning, logistics, transportation planning, and the traveling salesman problem. He was a masterful individual with superb insight, ingenuity, creativity, and elegance.

Bio:  Dr. Magnanti has spent his entire academic career at MIT where he is an Institute Professor and was formerly Dean of the School of Engineering and head of the Management Science area of the MIT Sloan School of Management among many other leadership roles. Much of his professional career has been devoted to education that combines engineering and management and to teaching and research in applied and theoretical aspects of large-scale optimization in areas such as production planning and scheduling, transportation planning, facility location, networks, logistics, and communication systems design. He is an inaugural Fellow and past president of INFORMS, past president of ORSA, a member of the National Academy of Engineering, and the American Academy of Arts and Sciences. His awards include the Lanchester Prize (1993) and the George E. Kimball Medal (1994) and awards at MIT for educational innovations and distinguished service. He has served on the advisory boards and board of directors for several leading universities and corporations.

Spring 2024: Data Science Distinguished Lecture Series

Wednesday, April10
Bodhi Sen, Columbia

Extending the Scope of Nonparametric Empirical Bayes

Co-sponsored with the Department of Statistics and Data Science

Abstract: In this talk we will describe two applications of empirical Bayes (EB) methodology. EB procedures estimate the prior probability distribution in a latent variable model or Bayesian model from the data. In the first part we study the (Gaussian) signal plus noise model with multivariate, heteroscedastic errors. This model arises in many large-scale denoising problems (e.g., in astronomy). We consider the nonparametric maximum likelihood estimator (NPMLE) in this setting. We study the characterization, uniqueness, and computation of the NPMLE which estimates the unknown (arbitrary) prior by solving an infinite-dimensional convex optimization problem. The EB posterior means based on the NPMLE have low regret, meaning they closely target the oracle posterior means one would compute with the true prior in hand. We demonstrate the adaptive and near-optimal properties of the NPMLE for density estimation, denoising and deconvolution.

In the second half of the talk, we consider the problem of Bayesian high dimensional regression where the regression coefficients are drawn i.i.d. from an unknown prior. To estimate this prior distribution, we propose and study a “variational empirical Bayes” approach — it combines EB inference with a variational approximation (VA). The idea is to approximate the intractable marginal log-likelihood of the response vector — also known as the “evidence” — by the evidence lower bound (ELBO) obtained from a naive mean field (NMF) approximation. We then maximize this lower bound over a suitable class of prior distributions in a computationally feasible way. We show that the marginal log-likelihood function can be (uniformly) approximated by its mean field counterpart. More importantly, under suitable conditions, we establish that this strategy leads to consistent approximation of the true posterior and provides asymptotically valid posterior inference for the regression coefficients.

Bio: Bodhi Sen is a Professor of Statistics at Columbia University, New York. He completed his Ph.D in Statistics from University of Michigan, Ann Arbor, in 2008. Prior to that, he was a student at the Indian Statistical Institute, Kolkata, where he received his Bachelors (2002) and Masters (2004) in Statistics. His core statistical research centers around nonparametrics — function estimation (with special emphasis on shape constrained estimation), theory of optimal transport and its applications to statistics, empirical Bayes procedures, kernel methods, likelihood and bootstrap based inference, etc. He is also actively involved in interdisciplinary research, especially in astronomy.

His honors include the NSF CAREER award (2012), and the Young Statistical Scientist Award (YSSA) in the Theory and Methods category from the International Indian Statistical Association (IISA). He is an elected fellow of the Institute of Mathematical Statistics (IMS).

Thursday, March 7
Jelena Bradic, UC San Diego

Exploring Robustness: Bridging Theory and Practice for Enhanced Discovery

Co-sponsored with the Department of Statistics and Data Science

ABSTRACT  This talk seeks to redefine the boundaries of statistical robustness. For too long, the field has languished in the shadows of contamination models, adversarial constructs, and outlier management—approaches that, while foundational, scarcely scratch the surface of potential that model misspecification offers. Our research reveals a fundamental link between robustness and causality, initiating an innovative era in data science. This era is defined by how causality enhances robustness, and in turn, how effectively applied robustness opens up unprecedented opportunities for scientific exploration.

BIO  Jelena Bradic is a Professor at the University of California, San Diego, where she specializes in statistics and data science within the Department of Mathematics and the Halicioglu Data Science Institute. Her research focuses on developing robust statistical methods that are resistant to model misspecification, with particular emphasis on high-dimensional data analysis, causal inference, and machine learning applications. Bradic holds a Ph.D. from Princeton University and has made significant contributions to the field of statistics. She is the co-editor in Chief of the first interdisciplinary journal between ACM and IMS named ACM/IMS Journal of Data Science, and  the recipient of several prestigious awards, including the  Wijsman Lecture (2023), a Discussion Paper in the Journal of the American Statistical Association (2020) and a Hellman Fellowship, recognizing her as a leading figure in statistical science.

Abstract: Geometry arises in a myriad ways across the sciences, and quite naturally also within AI and  optimization. In this talk I wish to share with you examples where geometry helps us understand problems in machine learning, optimization, and sampling. For instance, when sampling from densities supported on a manifold, understanding geometry and the impact of curvature are crucial; surprisingly, progress on geometric sampling theory helps us understand certain generalization properties of SGD for deep-learning! Another fascinating viewpoint afforded by geometry is in non-convex optimization: geometry can either help us make training algorithms more practical (e.g., in deep learning), it can reveal tractability despite non-convexity (e.g., via geodesically convex optimization), or it can simply help us understand important ideas better (e.g., eigenvectors, LLM training, etc.).

Ultimately, my hope is to offer the audience insights into geometric thinking, and to share with them some new tools that can help us make progress on modeling, algorithms, and applications. To make my discussion concrete, I will recall a few foundational results arising from our research, provide several examples, and note some open problems. 

Bio: Suvrit Sra is an Alexander von Humboldt Professor of AI at the Technical University of Munich (Germany), and an Associate Professor of EECS at MIT (USA), where he is also a member of the Laboratory for Information and Decision Systems (LIDS) and of the Institute for Data, Systems, and Society (IDSS). He obtained his PhD in Computer Science from the University of Texas at Austin. Before TUM & MIT, he was a Senior Research Scientist at the Max Planck Institute for Intelligent Systems, Tübingen, Germany. He has held visiting positions at UC Berkeley (EECS) and Carnegie Mellon University (Machine Learning Department) during 2013-2014. His research bridges mathematical topics such as differential geometry, matrix analysis, convex analysis, probability theory, and optimization with AI. He founded the OPT (Optimization for Machine Learning) series of workshops, held from OPT2008–2017 at the NeurIPS  conference. He has co-edited a book with the same name (MIT Press, 2011). He is also a co-founder and Chief Scientist of Pendulum, a global “Decision Aware AI” startup.

Tuesday, February 6
Stefanie Jegelka, MIT

Learning with Structure, Graphs, Geometry and Generalization 

co-sponsored with the Department of Computer Science

ABSTRACT  One grand goal of machine learning is to design widely applicable and resource-efficient learning models that are robust, even under commonly occurring distribution shifts in the input data. A promising step towards this goal is to understand and exploit “structure”, in the input data, latent space, model architecture and output. In this talk, I will illustrate examples of exploiting different types of structure in two main areas: representation learning on graphs and learning with symmetries.

First, graph representation learning has found numerous applications, including drug and materials design, traffic and weather forecasting, recommender systems and chip design. In many of these applications, it is important to understand (and improve) the robustness of relevant deep learning models. Here, we will focus on approaches to estimate and understand out-of-distribution predictions on graphs, e.g., training on small graphs and testing on large graphs with different degrees.

Second, in many applications, e.g. in chemistry, physics, biology or robotics, the data have important symmetries. Modeling such symmetries can help data efficiency and robustness of a model. Here, we will see an example of such a modeling task — neural networks on eigenvectors — and its benefits. Moreover, a formal, general analysis quantifies how symmetries improve data efficiency.

BIO  Stefanie Jegelka is a Humboldt Professor at TU Munich and an Associate Professor in the Department of EECS at MIT. Before joining MIT, she was a postdoctoral researcher at UC Berkeley, and obtained her PhD from ETH Zurich and the Max Planck Institute for Intelligent Systems. Stefanie has received a Sloan Research Fellowship, an NSF CAREER Award, a DARPA Young Faculty Award, the German Pattern Recognition Award, a Best Paper Award at ICML and an invitation as sectional lecturer at the International Congress of Mathematicians. She has co-organized multiple workshops on (discrete) optimization in machine learning and graph representation learning, and has served as an Action Editor at JMLR and a program chair of ICML 2022. Her research interests span the theory and practice of algorithmic machine learning, in particular, learning problems that involve combinatorial, algebraic or geometric structure.

Monday, February 5
Moon Duchin, Tufts University

 UNIVERSITY LECTURE

Algorithms, Race, and Redistricting: Can Computers Find Fairness?

This is a University Lecture and is co-sponsored  with the Center for Data Science for Enterprise and Society and the Jeb E. Brooks School of Public Policy

Today’s Supreme Court is unmistakably inclined to reject the use of race-conscious measures in law and policy — as Chief Justice Roberts memorably put it, “The way to stop discrimination on the basis of race is to stop discriminating on the basis of race.” The 2023 term saw high-profile challenges to the use of race data in college admissions and in political redistricting. On the gerrymandering front, the state of Alabama asked the court to adopt a novel standard using algorithms to certify race-neutrality, on the principle that computers don’t know what you don’t tell them. But do blind approaches find fairness? In this talk, Professor Duchin will review the very interesting developments of the last few decades — and the last few months! — on algorithms, race, and redistricting.

BIO  Moon Duchin is a Professor of Mathematics and a Senior Fellow in the Tisch College of Civic Life at Tufts University. Her pure mathematical work is in geometry, topology, groups, and dynamics, while her data science work includes collaborations in civil rights, political science, law, geography, and public policy on large-scale projects in elections and redistricting. She has recently served as an expert in redistricting litigation in Wisconsin, North Carolina, Alabama, South Carolina, Pennsylvania, Texas, and Georgia.  Her work has been recognized with an NSF CAREER grant, a Guggenheim Fellowship, and a Radcliffe Fellowship, and she is a Fellow of the American Mathematical Society. 

Friday, February 2
Moon Duchin, Tufts University

A Systems View of Fair Elections

This talk is part of the CAM Colloquium and Cornell’s Center for Data Science for Enterprise and Society’s Data Science Distinguished Lecture Series

ABSTRACT: The mathematical attention to voting systems has come in waves: the first wave was bound up with the development of probability, the second wave was axiomatic, and the third wave is computational. There are many beautiful results giving guarantees and obstructions when it comes to the provable properties of systems of election. But the axioms and objectives from this body of work are not a great match for the practical challenges of 21st century democracy. I will discuss ideas for bringing the tools of modeling and computation into closer conversation with the concerns of policymakers and reformers in the voting rights sphere. In particular, I’ll take a close look at ranked choice voting (or, as Politico recently called it, “the hottest political reform of the moment”).

Fall 2023: Data Science Distinguished Lecture Series

Spring 2023: Data Science Distinguished Lecture Series

Friday, March 10
Kristian Lum
University of Chicago Data Science Institute

Defining, Measuring, and Reducing Algorithmic Amplification

Co-sponsored with the AI Seminar and IS Colloquium

BIO Kristian Lum is an Associate Research Professor at the University of Chicago Data Science Institute. Previously, she was a Sr. Staff Machine Learning Researcher at Twitter where she led research as part of the Machine Learning Ethics, Transparency, and Accountability (META) team. She is a founding member of the ACM Conference on Fairness, Accountability, and Transparency and has served in various leadership roles since its inception, growing this community of scholars and practitioners who care about the responsible use of machine learning systems, and she is a recent recipient of the COPSS Emerging Leaders Award and NSF Kavli Fellow. Her research looks into (un)fairness of predictive models with particular attention to those used in a criminal justice setting.

ABSTRACT  As people consume more content delivered by recommender systems, it has become increasingly important to understand how content is amplified by these recommendations. Much of the dialogue around algorithmic amplification implies that the algorithm is a single machine learning model acting on a neutrally defined, immutable corpus of content to be recommended. However, there are several other components of the system that are not traditionally considered part of the algorithm that influence what ends up on a user’s content feed. In this talk, I will enumerate some of these components that influence algorithmic amplification and discuss how these components can contribute to amplification and simultaneously confound its measurement. I will then discuss several proposals for mitigating unwanted amplification, even when it is difficult to measure precisely.

Wednesday, May 10 – 3:30 – 4:30 pm, Gates 114
Jessica Hullman
Northwestern University

Toward Robust Data Visualization for Inference

Co-sponsored with the IS Colloquium

BIO Dr. Jessica Hullman is the Ginni Rometty Associate Professor of Computer Science at Northwestern University. Her research addresses challenges that arise when people draw inductive inferences from data interfaces. Hullman’s work has contributed visualization techniques, applications, and evaluative frameworks for improving data-driven inference in applications like visual data analysis, data communication, privacy budget setting, and responsive design. Her current interests include theoretical frameworks for formalizing and evaluate the value of a better interface and elicitation of domain knowledge for data analysis. Hullman’s work has been awarded best paper awards at top visualization and HCI venues. She is the recipient of a Microsoft Faculty Fellowship (2019) and NSF CAREER, Medium, and Small awards as PI, among others.

ABSTRACT  Research and development in computer science and statistics have produced increasingly sophisticated software interfaces for interactive visual data analysis, and data visualizations have become ubiquitous in communicative contexts like news and scientific publishing. However, despite these successes, our understanding of how to design robust visualizations for data-driven inference remains limited. For example, designing visualizations to maximize perceptual accuracy and users’ reported satisfaction can lead people to adopt visualizations that promote overconfident interpretations. Design philosophies that emphasize data exploration and hypothesis generation over other phases of analysis can encourage pattern-finding over sensitivity analysis and quantification of uncertainty. I will motivate alternative objectives for measuring the value of a visualization, and describe design approaches that better satisfy these objectives. I will discuss how the concept of a model check can help bridge traditionally exploratory and confirmatory activities, and suggest new directions for software and empirical research.

Fall 2022: Data Science Distinguished Lecture Series

November 15, 2022 @ 4:15 : Center for Data Science for Enterprise & Society,  Data Science Distinguished Lecture Series

Nonlinear Optimization in the Presence of Noise

This talk is co-sponsored with the School of Operations Research and Information Engineering

Speaker: Jorge Nocedal

Date: Tuesday, November 15, 2022

Time/Location:
4:15 p.m. , Rhodes Hall, Room 253
Reception 3:45 p.m., Rhodes 258

Abstract:  We begin by presenting three case studies that illustrate the nature of noisy optimization problems arising in practice. They originate in atmospheric sciences, machine learning, and engineering design. We wish to understand the source of the noise (e.g. a lower fidelity model, sampling or reduced precision arithmetic), its properties, and how to estimate it. This sets the stage for the presentation of our goal of redesigning constrained and unconstrained nonlinear optimization methods to achieve noise tolerance.

Bio:  Jorge Nocedal is the Walter P. Murphy Professor in the Department of Industrial Engineering and Management Sciences at Northwestern University. He studied at UNAM (Mexico) and Rice University. His research is in optimization, both deterministic and stochastic, with emphasis on very large-scale problems.  He is a SIAM Fellow, was awarded the 2012 George B. Dantzig Prize and the 2017 Von Neumann Theory Prize, for contributions to theory and algorithms of nonlinear optimization. He is a member of the US National Academy of Engineering.