webnotice

Seminar by Tim Swartz

Greg Preston — Wed, 02 Jul 2025 18:02:32 +0000

Seminar by Tim Swartz Greg Preston Wed, 07/02/2025 - 14:02

Sports Analytics Club seminar

Tim Swartz
Simon Fraser University

Room: M3 3127

Two Problems in Soccer Analytics

This talk concerns two problems in soccer analytics that both rely on tracking data. The first problem begins with a review of average aging curves in sport. Then, a new approach is introduced which addresses personal aging curves in soccer, an essential problem of interest which has not been previously addressed.

The second problem concerns the development of a metric that identifies soccer players who have a similar style to a player of interest. Whereas performance variables have been well studied, the same is not true of stylistic variables. Unlike assessments from scouting, the metric is automatic and objective. The metric is developed using a Bayesian framework.

Biography:

Tim Swartz is Professor and former Chair in the Department of Statistics and Actuarial Science at Simon Fraser University. He obtained a PhD and MSc in Statistics from the University of Toronto and a BMath from the University of ��ݮ��Ƶ. He has over 120 research publications and has written several books including an Oxford text (2000) with Michael Evans on Approximating Integrals via Monte Carlo and Deterministic Methods. He is Fellow of the American Statistical Association and is AE for five journals. Most of his current research involves sports analytics.

Seminar by Peng Shi

Greg Preston — Tue, 10 Jun 2025 17:11:38 +0000

Seminar by Peng Shi Greg Preston Tue, 06/10/2025 - 13:11

Actuarial Science and Financial Mathematics seminar series

Peng Shi
University of Wisconsin-Madison

Room: M3 3127

Actuarial Modeling and Pricing of Deductible Insurance Contracts

Insurers use experience rating to adjust premiums for renewing policyholders based on their claims history. This paper focuses on experience rating for insurance contracts with deductibles, which policyholders select endogenously as a cost-sharing tool to reduce inefficiencies from information asymmetry.

We propose a copula-based longitudinal model for repeated risk outcome measurements, addressing the endogeneity of deductible choices. Our framework relaxes the common assumption that deductibles are exogenous, aligning with economic theories of adverse selection and moral hazard. It supports various risk outcome distributions, making it well-suited for predictive experience rating. Applied to commercial property insurance data, our method reveals a negative relationship between deductible levels and underlying risk. It enables improved risk segmentation and identification of profitable policies, outperforming traditional models that treat deductibles as exogenous.

Seminar by Bin Zou

Greg Preston — Tue, 27 May 2025 13:15:13 +0000

Seminar by Bin Zou Greg Preston Tue, 05/27/2025 - 09:15

Actuarial Science and Financial Mathematics seminar series

Bin Zou
University of Connecticut

Room: M3 3127

Optimal Proportional Insurance under Claim Habit

In this paper, we study a two-period optimal insurance problem for a policyholder with mean-variance preferences who purchases proportional insurance at the beginning of each period. The insurance premium is calculated by a variance premium principle with a risk loading that depends on the policyholder’s claim history. We derive the time-consistent optimal insurance strategy in closed form and the optimal constant precommitment strategy in semiclosed form. For the optimal general precommitment strategy, we obtain the solution for the second period semi-explicitly and, then, the solution for the first period numerically via an efficient algorithm.

Furthermore, we compare the three types of optimal strategies, highlighting their differences, and we examine the impact of the key model parameters on the optimal strategies and value functions.

Seminar by Michael Kupper

Greg Preston — Thu, 01 May 2025 12:40:42 +0000

Seminar by Michael Kupper Greg Preston Thu, 05/01/2025 - 08:40

Actuarial Science and Financial Mathematics seminar series

Michael Kupper
University of Konstanz

Room: M3 3127

Risk measures based on weak optimal transport and approximation of drift control problems

We discuss convex risk measures with weak optimal transport penalties and show that these risk measures admit an explicit representation via a nonlinear transform of the loss function. We discuss several examples, including classical optimal transport penalties and martingale constraints. In the second part of the talk, we focus on the composition of related functionals. We consider a stochastic version of the Hopf–Lax formula, where the Hopf–Lax operator is composed with the transition kernel of a Lévy process. We show that, depending on the order of composition, one obtains upper and lower bounds for the value function of a stochastic optimal control problem associated with drift-controlled Lévy dynamics. The value function of the control problem is approximated both from above and below as the number of iterations tends to infinity, and we provide explicit convergence rates for the approximation procedure.

The talk is based on joint work with Max Nendel and Alessandro Sgarabottolo.

Seminar by Marcel Nutz

Greg Preston — Tue, 29 Apr 2025 19:29:31 +0000

Seminar by Marcel Nutz Greg Preston Tue, 04/29/2025 - 15:29

CANSSI Ontario STatistics Seminars (CAST)

Marcel Nutz
Columbia University

Room: M3 3127

This is a free event, but please register for this talk on CANSSI's website.

Sparse Regularized Optimal Transport

Entropic optimal transport — the optimal transport problem regularized by KL divergence — is highly successful in statistical applications. Thanks to the smoothness of the entropic coupling, its sample complexity avoids the curse of dimensionality suffered by unregularized optimal transport. The flip side of smoothness is overspreading: the entropic coupling always has full support, whereas the unregularized coupling that it approximates is usually sparse, even given by a map. Regularizing optimal transport by less-smooth f-divergences such as Tsallis divergence is known to allow for sparse approximations, but is often thought to suffer from the curse of dimensionality as the couplings have limited differentiability and the dual is not strongly concave. We refute this conventional wisdom and show, for a broad family of divergences, that the key empirical quantities converge at the parametric rate, independently of the dimension. More precisely, we provide central limit theorems for the optimal cost, the optimal coupling, and the dual potentials induced by i.i.d. samples from the marginals. (Joint work with Alberto Gonzalez-Sanz and Stephan Eckstein.)

Seminar by Saeed Ghadimi

Greg Preston — Thu, 24 Apr 2025 14:35:04 +0000

Seminar by Saeed Ghadimi Greg Preston Thu, 04/24/2025 - 10:35

Probability seminar series

Saeed Ghadimi
University of ��ݮ��Ƶ

Room: M3 3127

An Adversarially Robust Formulation of Linear Regression with Missing Data

In this talk, I'll present a robust framework to perform linear regression with missing entries in the features. By considering an elliptical data distribution, and specifically a multivariate normal model, we are able to conditionally formulate a distribution for the missing entries and present a robust framework, which minimizes the worst-case error caused by the uncertainty in the missing data. The proposed formulation ultimately reduces to a convex program, for which we develop a customized and scalable solver. I'll also discuss the consistency and structural behavior of the proposed framework in asymptotic regimes. Finally, I'll present some numerical experiments performed on synthetic, semi-synthetic, and real data, and show how the proposed formulation improves the prediction accuracy and robustness.

Seminar by David Awosoga

Greg Preston — Wed, 09 Apr 2025 12:58:49 +0000

Seminar by David Awosoga Greg Preston Wed, 04/09/2025 - 08:58

Student seminar series

David Awosoga
PhD Candidate, University of ��ݮ��Ƶ

Room: M3 3127

Applied Bayesian Reinforcement Learning for Credit Assignment in Volleyball

Understanding individual contribution towards overall group output, otherwise known as credit assignment, is a complex and multifaceted area of study relevant across a wide domain of fields. Various methods have been proposed to quantify contribution based on recorded actions and active participation, but an understudied area of credit assignment is in investigating the optimality of individual decision-making. One area of application is in team sports, where effectively assigning credit to players allows coaches to optimally construct rosters, allocate playing time, and improve upon strategic and tactical considerations. This work focuses on volleyball, where the most influential decisions are those made by the “setter”, who is responsible for distributing attack opportunities among their teammates. Drawing inspiration from Bayesian reinforcement learning strategies for addressing sequential decision-making under uncertainty, this talk will propose an improved setter evaluation framework.

Seminar by Weijing Tang

Greg Preston — Tue, 01 Apr 2025 18:00:54 +0000

Seminar by Weijing Tang Greg Preston Tue, 04/01/2025 - 14:00

Statistics and Biostatistics seminar series

Weijing Tang
Carnegie Mellon University

Held virtually on Zoom Meetings.

Watch the talk in M3 3127.

Inference and Learning for Signed Networks Guided by Social Theory

In many real-world networks, relationships often go beyond simple presence or absence; they can be positive (e.g., friendship, alliance, and mutualism) or negative (e.g., enmity, disputes, and competition). These negative relationships display substantially different properties from positive ones, and more importantly, their presence interacts in unique ways. The balance theory originating from social psychology, illustrated by proverbs like "a friend of my friend is my friend'' and "an enemy of my enemy is my friend'', provides insight into the formation mechanism of positive and negative connections. In this talk, we characterize the balance theory with a novel and natural notion of population-level balance. We propose a nonparametric inference method to evaluate the real-world evidence of population-level balance in signed networks. Inspired by the empirical findings, we further develop latent variable models for signed networks that accommodate the balance theory for embedding learning and community detection.

Seminar by Venu Veeravalli

Greg Preston — Tue, 01 Apr 2025 17:25:31 +0000

Seminar by Venu Veeravalli Greg Preston Tue, 04/01/2025 - 13:25

Statistics and Biostatistics seminar series

Jointly held with Applied Mathematics

Venu Veeravalli
University of Illinois at Urbana-Champaign

Room: M3 3127

Out-of-Distribution Detection via Multiple Testing

Out-of-Distribution (OOD) detection in machine learning refers to the problem of detecting whether the machine learning model's output can be trusted at inference time. This problem has been described qualitatively in the literature, and a number of ad hoc tests for OOD detection have been proposed. In this talk we outline a principled approach to the OOD detection problem, by first defining the problem through a hypothesis test that includes both the input distribution and the learning algorithm. Our definition provides insights for the construction of good tests for OOD detection. We then propose a multiple testing inspired procedure to systematically combine any number of different OOD test statistics using conformal p-values. Our approach allows us to provide strong guarantees on the probability of incorrectly classifying an in-distribution sample as OOD. In our experiments, we find that the tests proposed in prior work perform well in specific settings, but not uniformly well across different types of OOD instances. In contrast, our proposed method that combines multiple test statistics performs uniformly well across different datasets, neural networks and OOD instances. We will end the talk with a discussion of the application of the multiple testing approach to the problem of hallucination detection in LLMs.

Seminar by Max Nendel

Greg Preston — Fri, 28 Mar 2025 18:23:54 +0000

Seminar by Max Nendel Greg Preston Fri, 03/28/2025 - 14:23

Student seminar series

Max Nendel
Associate Professor, University of ��ݮ��Ƶ

Room: M3 3127

Upper Comonotonicity and Risk Aggregation under Dependence Uncertainty

In this talk, we study dependence uncertainty and the resulting effects on tail risk measures, which play a fundamental role in modern risk management. We introduce the notion of a regular dependence measure, defined on multi-marginal couplings, as a generalization of well-known correlation statistics such as the Pearson correlation. The first main result states that even an arbitrarily small positive dependence between losses can result in perfectly correlated tails beyond a certain threshold and seemingly complete independence before this threshold. In a second step, we focus on the aggregation of individual risks with known marginal distributions by means of arbitrary nondecreasing left-continuous aggregation functions. In this context, we show that under an arbitrarily small positive dependence, the tail risk of the aggregate loss might coincide with the one of perfectly correlated losses. A similar result is derived for expectiles under mild conditions. In a last step, we discuss our results in the context of credit risk, analyzing the potential effects on the value at risk for weighted sums of Bernoulli distributed losses. The talk is based on joint work with Corrado De Vecchi and Jan Streicher.