Schedule for Fall 2018
Seminars are on Wednesdays
Time: 12:00pm – 1:00pm
Location: Room 1025, 1255 Amsterdam Avenue
Contacts: Yuling Yao, Owen Ward
Information for speakers: For information about schedule, direction, equipment, reimbursement and hotel, please click here.
Prof. Richard Davis, Prof. Bodhi Sen and Andrew Davison (Columbia)
“New Students Welcome and Introductions”
Wenda Zhou, Florian Stebegg, Shuaiwen Wang, Chengliang Tang, and others (Columbia)
“Students Summer internship”
Two Sigma “Two Sigma Info Session”
Susanna Makela (Google)
|10/3/18||Ari Brill (Columbia, Physics) “Deep Learning for Gamma Ray Astronomy”|
|10/10/18||SIG Quant Info Session|
|10/17/18||Victor Veitch (Columbia)|
Promit Ghosal (Columbia University)
Title: Monge-Kantorovich ranks and quantiles: Definitions and hypothesis testing.
Abstract: Ranks and quantiles are two important tools in the nonparametric statistics. There were enormous progress in one dimensional rank based inference in the past. Over the last three decades, several new notions of multidimensional ranks and quantiles are proposed. Recently, Chernozhukov et al. (2015) introduced Monge-Kantorovich (MK) ranks and quantiles whose denitions are motivated by the optimal transportation theory. In an ongoing work with Prof. Bodhi Sen, we propose an one sample and a two sample test for testing the equality of multidimensional distributions based on the MK ranks and quantiles. This talk is the first one in a series of two talks on this joint work. The second talk will be given by Prof. Bodhi Sen next week. In this week’s talk, we will primarily focus on the denitions of relevant objects and some previous works. If time permits, we will discuss the consistency and other properties of our tests and show an outline of the pointwise rate of convergence result of the MK quantile function. No prior knowledge of the optimal transport theory will be assumed.
Prof. Bodhi Sen (Columbia University)
Prof. Sumit Mukherjee (Columbia University)
“Limit theory for permutations”
Permutation limit theory first originated in Combinatorics, and is very much motivated by (dense) graph limit theory. We will first give examples of models of random permutations (which includes the famous Mallows models) for which we know the existence of a limit. As an application of this, we will compute limiting properties of various statistics on permutations, such as the number of fixed points, number of cycles of a given length, and number of inversions. As another application of this theory, we will compute asymptotics of the log normalizing constant for some exponential families on the space of permutations, which in turn will be used to show existence of consistent estimators in these models.
|11/21/18||No Seminar – Academic Holiday|
Elizabeth Ogburn (Johns Hopkins)
“Toward valid causal and statistical inference with social network data”
Interest in and availability of social network data has led to increasing attempts to make causal and statistical inferences using data collected from subjects linked by social network ties. When social relations can engender dependence in the variables of interest, treating such observations as independent results in invalid, anti-conservative statistical inference, but there is a dearth of methods that can account for this kind of dependence. We develop a test for network dependence that can be used to screen for the appropriateness of i.i.d. statistical methods and apply it to data from the Framingham Heart Study. Our results suggest that some of the many decades worth of research on coronary heart disease and other health outcomes using FHS data could be invalid due to unacknowledged network dependence. We also extend recent work by van der Laan (2014) on causal inference for causally connected units to more general social network settings: we describe estimation and inference for causal effects that are specifically of interest in social network settings, and our asymptotic results allow for dependence of each observation on a growing number of other units as sample size increases.
Prof. Victor de la Pena (Columbia University)
“Beyond Martingales: Decoupling and Self-Normalization”
Abstract: In this talk I will provide an overview of the theory and applications of decoupling and self-normalization. These inter-related areas grew out of the need to extend martingale methods to high and infinite dimensions (Banach Spaces), and complex non-linear dependence structures. Decoupling provides tools for symmetrizing as well as treating dependent variables as if they were independent. A prototypical example of a self-normalized process is the t-statistic. The tools developed have successfully applied in diverse areas such as the the theory of empirical processes for dependent variables, including U-processes, on-line learning, classification algorithms, optimization and several others.
Prof. Yang Feng (Columbia University)
“Are there any community structure in a hypergraph?”