Past Seminar Series

Date: 2026-10-23

Time: 15:30-16:30 (Montreal time)

Location: Online via Zoom

https://mcgill.zoom.us/j/83173904340

Meeting ID: 831 7390 4340

Abstract:

Targets of inference in nonparametric instrumental variables, proximal causal inference, and shadow-variable models are often continuous linear functionals of solutions to a pair of ill-posed inverse problems. Recent work has shown that, under minimal assumptions, ill-posedness does not preclude pointwise root-n inference. We show that pointwise guarantees need not translate into locally reliable inference. In a baseline model sufficient for a Neyman orthogonal estimating equation to exist, the target is Hellinger discontinuous, which rules out locally uniformly valid inference. A stronger model that uniformly bounds the minimum-norm solution to one of the inverse problems restores continuity. However, pathwise differentiability then holds only if that solution is unique at the true data-generating process. When it fails, Wald-type confidence intervals from standard procedures undercover asymptotically under some local alternatives. Pathwise differentiability is also restored, even without uniqueness, under a uniform bound on a beta-source norm. We relate these regimes to very weak, weak, and strong identification in linear IV models.

Date: 2026-09-25

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/89492447923

Meeting ID: 894 9244 7923

Abstract:

While classical statistical methods are designed for testing hypotheses about pre-specified models, the reality of modern science is that analysts often explore their data before coming up with models and hypotheses of interest. We refer to the practice of using the same data to generate and then test a hypothesis as double dipping. Classical statistical hypothesis tests will fail to control the Type 1 error rate in settings that involve double dipping. Often, we avoid double dipping by splitting our observations into a training set and a test set. While this sample splitting approach is straightforward and easy to understand, it is generally inapplicable in unsupervised settings. Motivated by unsupervised problems that arise in the analysis of single-cell RNA sequencing data, we first propose data thinning, an alternative to sample splitting that splits each observation in a dataset into two independent pieces. We show that this method provides an elegant solution to our motivating problems under distributional assumptions. We then show how the data fission framework of Leiner et al. (2025) can be fruitfully applied to circumvent the distributional assumptions of data thinning. We conclude by discussing the promise and the challenges of data fission in the context of inference after clustering and inference after variable selection in logistic regression.

Date: 2026-04-17

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/86146204241

Meeting ID: 861 4620 4241

Passcode: None

Abstract:

This talk presents two complementary methodological frameworks for improving single-cell CRISPR screens, from both the analysis and study-design perspectives. The first, spaCRT, addresses a key inferential challenge in single-cell data: gene expression measurements are often sparse and noisy, so standard asymptotic tests can miscalibrate significance while resampling methods, though more reliable, are often too slow at scale. spaCRT overcomes this by using saddlepoint approximations to provide a closed-form approximation to the resampling p-value, yielding accurate error control, competitive power, and substantial computational savings. The second, PerturbPlan, addresses the design side of these experiments: because CRISPR screens are expensive, experimental choices with similar budgets can differ greatly in statistical power. PerturbPlan uses an analytic power formula, validated through simulations and real datasets, to provide near-instant power estimates and generate cost-aware, power-optimized designs across a broad range of common study settings. Together, these frameworks aim to make single-cell CRISPR studies both more statistically reliable and more efficiently designed.

Date: 2026-04-10

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/84830806641

Meeting ID: 848 3080 6641

Passcode: None

Abstract:

From a macroprudential view, systemic risk arises from the gradual buildup of financial imbalances across the system. In probabilistic terms, these vulnerabilities manifests in the extremal dependence structure of the system. In this talk, I present an extreme value framework for characterizing extremal dependence in multivariate distributions based on tail expansions of copulas. This framework yields a new approach to Conditional Value-at-Risk (CoVaR), one of the most widely used measures of systemic risk. Our work characterizes the possible tail regimes of CoVaR through the limiting behavior of the copula conditional distribution and proves that these regimes can be determined by the joint tail expansions of the copula. Building on this characterization, we also propose a minimum-distance estimation approach for CoVaR and establishes its asymptotic properties. The talk also features an empirical study of systemic risk in the U.S. market from 2000 to 2025. It shows how the proposed methodology can reveal changes in systemic risk, and help distinguish the systemic roles of different assets and institutions. The findings have useful implications for macroprudential surveillance and risk management.

Date: 2026-03-13

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/82187303482

Meeting ID: 821 8730 3482

Passcode: None

Abstract:

In this talk, we propose a copula-based framework for regression modelling in which the conditional distribution of the response variable, given covariates, is specified through a parametric family of continuous or discrete distributions. For mixed models, we incorporate cluster-level dependence by introducing a common latent factor modeled via a factor copula. We discuss the estimation of both the copula parameters and the marginal parameters, and we derive the asymptotic behavior of the resulting estimators. Numerical experiments are performed to assess the precision of the estimators for finite samples. An example of an application is given using COVID-19 vaccination hesitancy from several countries. This is a joint work with Pavel Krupskii and Bruno Remillard.

Date: 2026-02-20

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/82441217734

Meeting ID: 824 4121 7734

Passcode: None

Abstract:

Hierarchical clustering is one of the most widely used approaches for exploring data. However, its greedy nature makes it highly sensitive to small perturbations, blurring the lines between genuine structure and spurious patterns.  In this work, we show how randomizing hierarchical clustering can be useful not just for assessing clustering stability but also for designing valid hypothesis testing procedures based on clustering results.  In particular, we propose a method for constructing a valid p-value at each node of the hierarchical clustering dendrogram that quantifies evidence against performing the greedy merge.  Furthermore, we show how our p-values can be used to estimate the number of clusters, with a probabilistic guarantee on overestimation of the number of clusters.  This is joint work with Di Wu and Snigdha Panigrahi.

Date: 2026-02-13

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/81379129957

Meeting ID: 813 7912 9957

Passcode: None

Abstract:

Distorting multivariate distributions is a useful approach for introducing flexibility and capturing model uncertainty. In particular, applying distortions to the copulas representing the underlying dependence structure allows one to generate new, flexible dependence models from existing ones. In this presentation, we investigate the extremal domain of attraction problem for Morillas-type distorted copulas. We establish not only conditions under which such copula-to-copula transformations alter the respective asymptotic behavior, but also discuss conditions under which the distorted copulas remain in the same domain of attraction as the initial undistorted copula. Furthermore, we discuss the effect of these distortions on multivariate risk measures, such as the lower-orthant Value-at-Risk and Range-Value-at-Risk. Finally, we propose a simulation algorithm for Morillas-type distorted copulas, addressing a gap in the literature and providing the means to utilize these modified dependence structures in practice. We end the presentation with an application of distorted copula models for hail insurance.

Date: 2026-01-23

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/82195728045

Meeting ID: 821 9572 8045

Passcode: None

Abstract:

The analysis of extreme wave surge heights in Atlantic Canada is key in determining areas that are subject to flooding or at risk of severe damage from intense storms. One method for modelling extreme events is through the block maxima approach which divides a series of observations into equal-sized blocks to extract the maxima after which inference is conducted on the generalized extreme value (GEV) distribution. When observations at the series level are missing, the observed block maxima may not correspond to the true block maxima. In this presentation, we introduce this missing data problem in the context of an extreme value analysis and explain how concepts from survival analysis can be used to improve inferences on the GEV distribution using the observed block maxima.

Date: 2026-01-16

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/89692052783

Meeting ID: 896 9205 2783

Passcode: None

Abstract:

Clustering is a fundamental tool for uncovering heterogeneity in data, yet a longstanding challenge lies in assessing whether detected clusters represent genuine structure or arise from sampling variability, and in determining which variables drive the clustering structure. Statistical significance clustering (SigClust; Liu et al. (2008)) addresses the first challenge by testing the cluster index under a Gaussian null, estimating its distribution via Monte Carlo simulation in high dimensions. We propose SigClust-DE, which builds on recent advances in high-dimensional covariance estimation to improve the accuracy of SigClust and extends it to variable-level inference. In particular, SigClust-DE unifies clustering significance testing and differential expression (DE) analysis, a central task in RNA-seq studies. By leveraging the Monte Carlo framework, our method controls type I error while maintaining high power for variable selection. Through extensive simulations and an application to RNA-seq data, we show that SigClust-DE achieves more accurate covariance estimation, effectively controls false discoveries, and substantially improves power in detecting differentially expressed variables, providing a general framework for clustering significance and variable-level inference in high-dimensional data.

Date: 2025-12-05

Time: 15:30-16:30 (Montreal time)

Location: In person, Burnside 1104

https://mcgill.zoom.us/j/83026954715

Meeting ID: 830 2695 4715

Passcode: None

Abstract:

Shannon’s entropy is a cornerstone of information theory, quantifying uncertainty within a probability distribution. However, the classical definition may fail for distributions with heavy tails or infinite alphabets, leaving gaps in its theoretical foundation. This talk introduces a framework called Generalized Shannon’s Entropy (GSE), which extends the original concept to ensure well-definedness and robustness under broader conditions.