To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: The study explores the identification of strongly lensed galaxies, rare astronomical phenomena where massive objects bend light from distant sources, creating magnified and distorted images. Strongly lensed galaxies are crucial in astronomy as they provide insights into dark matter distribution, galaxy mass profiles, and cosmological parameters. To efficiently identify these lenses from vast datasets, images were ranked based on their probability of being strong lens candidates. A Learning-to-Rank (LTR) approach using Support Vector Machines (SVM) was implemented and compared with Convolutional Neural Networks (CNNs) and other classifiers. LTR with SVM demonstrated superior performance, achieving a high AUC and accuracy, outperforming CNNs in both efficiency and classification precision. This method facilitates efficient candidate selection, enhancing the potential for cosmological studies.

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: We develop the model-based latent causal socioeconomic health (LaCSH) index, with uncertainty bounds, at the national level. Extending the latent health factor index (LHFI) modeling approach to assess ecosystem health, LaCSH integratively models the hierarchical relationship among the nation’s societal health or well-being (latent / intangible), socio-economic metrics (e.g., GDP), the covariates that drive the notion of well-being (e.g., natural resources), and a continuous variable that reflects policy (e.g., government mandated maternity leave days). In addition to making statistical inference for socio-economic health, LaCSH facilitates the evaluation of potential causal impact of the policy on health. A formal spatial component in the LaCSH framework allows us to compare the socio-economic health of countries around the world based on various metrics, covariates, and two different policy variables that pertain to socio-economic well-being. This is joint work led by FS Kuh with AHWestveld (https://arxiv.org/abs/2009.12217)

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: Our Sun is dynamic with solar activity peaking approximately every 11 years. This rise in activity increases the chance of rare solar wind events (e.g. coronal mass ejections). The most recognizable effect of this rise in solar activity is the increase of planetary aurorae. We are currently approaching the peak of solar cycle 25 and in the last few months have observed several large coronal mass ejections arrive at Earth, spawning visible low-latitude aurora (including over Vancouver).

Spacecraft assets throughout the solar system observe our Sun and the solar wind. However, these datasets can only provide discontinuous spatiotemporal observations of a very large and dynamic system. Traditionally these observational assets are combined with high-fidelity physical models (e.g. magnetohydrodynamics). But these models are computationally expensive which limits the potential number of simulations. This bottleneck (sparse datasets, expensive forward physics-based models) is a ubiquitous challenge for inverse problems in the Earth and planetary sciences. In the case of Mars, this methodological bottleneck has limited our understanding of how the rise and fall of solar cycle activity affects planetary habitability.

In this presentation, I will pose estimating the solar wind during a recent rare solar wind event at Mars as an inverse problem. I will then discuss a Bayesian approach to this inverse problem using Gaussian processes as a low-fidelity emulator of a physics-based model and the scientific conclusions we are gaining about Mars from this effort. I will conclude with an outlook for simulation-based inference in Earth and planetary sciences.

Bio: Dr. Abigail (Abby) Azari is a Data Science Post-Doctoral Fellow in the Department of Earth, Ocean and Atmospheric Sciences where she works with Dr. Catherine Johnson (EOAS), Dr. Lindsey Heagy (EOAS), and Dr. Frank Wood (CS). Dr. Azari is a member of the NASA MAVEN Science Team; a spacecraft that has orbited Mars since 2014. Her research generally focuses on machine learning for scientific insights about planetary space environments. She was previously a postdoc at the UC Berkeley’s Space Sciences Lab. She received her Ph.D. in 2020 from the University of Michigan where she was an NSF Graduate Research Fellow and a NASA Earth and Space Sciences Fellow.

In January 2025, Dr. Azari will be joining the University of Alberta’s Physics and Electrical and Computer Engineering departments as an incoming faculty member and fellow of the Alberta Machine Intelligence Institute.

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: Hidden Markov models (HMMs) can infer biological rhythms by linking observations to underlying latent states of the biological process, often assuming state transition probabilities follow fixed, regular cycles. However, rhythms can fluctuate due to internal and external factors. I extend HMMs to model “irregular rhythms” that vary in frequency or stability over time. I analyze motor activity data from patients with depression to infer their circadian rhythms, which repeat every 24 hours but often exhibit irregularities. To jointly model state transition probabilities across all patients, accounting for daily behavioural cycles, daily variability, and individual variability, I formulate these probabilities to depend on time-of-day, day, and random effects. This approach provides insights into the regularities and trends of state-switching dynamics, revealing that transition probabilities do not always adhere to a regular daily cycle. Overall, this work advances the modelling of irregular rhythms in HMMs and contributes to a deeper understanding of circadian-related health issues.

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: The Gibbs sampler is a Markov Chain Monte Carlo algorithm that iteratively samples from the conditional distributions of a probability measure of interest and is widely used in computational statistics. Under the assumption of log-concavity, for its random scan version, we provide a sharp bound on the speed of convergence in relative entropy. Assuming that evaluating conditionals is cheap compared to evaluating the joint density, our results imply that the number of full evaluations required for the Gibbs sampler to mix grows linearly with the condition number and is independent of the dimension. This contrasts with gradient-based methods, whose mixing time typically increases with the dimension. Our techniques also allow us to analyze Metropolis-within-Gibbs schemes, as well as the Hit-and-Run algorithm. This is joint work with Filippo Ascolani and Giacomo Zanella. 

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: The work of the Flexible Learning in Statistics Group ranges from conducting studies of important aspects of statistics education to developing and testing resources for difficult statistics concepts. In this seminar, students will present several recent projects: using student focus groups to assess Shiny apps, developing and testing interactive resources to improve understanding of Bayesian inference, enhancing Stat 251 labs by creating active learning material and introducing pre-lab quizzes, and conducting a study of the impact of exam question wording on the performance of students with English as an Additional Language (EAL). You’ll also hear about StatEngage, the ASDa-led project to guide students through the challenges of consulting.

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: Many interventions in causal inference can be represented as transformations of the variables of interest. Abstracting interventions in this way allows us to identify a local symmetry property exhibited by many causal models under interventions. Where present, this symmetry can be characterized by a type of map called a cocycle, an object that is central to dynamical systems theory. We show that such cocycles exist under general conditions and are sufficient to identify interventional distributions and, under suitable assumptions, counterfactual distributions. We use these results to derive cocycle-based estimators for causal estimands and show that they achieve semiparametric efficiency under standard conditions. Since entire families of distributions can share the same cocycle, these estimators can make causal inference robust to mis-specification by sidestepping superfluous modelling assumptions. We demonstrate both robustness and state-of-the-art performance in several simulations, and apply our method to estimate the effects of 401(k) pension plan eligibility on asset accumulation using a real dataset.

Joint work with Hugh Dance (UCL/Gatsby Unit): https://arxiv.org/abs/2405.13844

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: The presence of big data, characterized by exceptionally large sample size, often brings the challenge of outliers and data distributions that exhibit heavy tails. An online learning estimation that incorporates anti-outlier capabilities while not relying on historical data is therefore urgently required to achieve robust and efficient estimators. In this talk, we introduce an innovative online learning approach based on a mode kernel-based objective function, specifically designed to address outliers and heavy-tailed distributions in the context of big data. The developed approach leverages mode regression within an online learning framework that operates on data subsets, which enables the continuous updating of historical data using pertinent information extracted from a new data subset. We demonstrate that the resulting estimator is asymptotically equivalent to the mode estimator calculated using the entire dataset. Monte Carlo simulations and an empirical study are presented to illustrate the finite sample performance of the proposed estimator.

To join this seminar virtually: Please request Zoom connection details from ea@stat.ubc.ca.

Abstract: This session will be a perspective of what working as a statistics consultant in a contract research organisation for pharmaceutical/biotech companies entails. In addition to an overview of potential career paths, the specific critical tasks and responsibilities involved for a statistician working in real-world data will be discussed.

A look into the type of statistical methodologies through case studies will be provided, demonstrating how they play a role in drug development, regulatory submissions, and health technology assessments. This sets the stage for the discussion of potential research collaborations between UBC students and industry, where students can have the opportunity to advance health research whilst gaining experience on whether a career in health research is of interest.