Seminars

Abstract :

Hawkes processes are concise mathematical representations of diverse point process data, ranging from disease spread and wildfire occurrences to non-physical phenomena such as financial asset price movements. Models for point process data are often fit using maximum likelihood (MLE) or Markov Chain Monte Carlo (MCMC), but such methods are slow or computationally intractable for data with large n. In this talk, I will present a novel estimator based on the Stoyan-Grabarnik (sum of inverse intensity) statistic. Unlike MLE or MCMC approaches, the proposed estimator does not require approximation of a computationally expensive integral. I will show that under quite general conditions, this estimator is consistent for estimating parameters governing spatial-temporal point processes such as the Hawkes process and present simulations demonstrating the performance of the estimator. In the second portion of the talk, I will discuss increasingly flexible parametric Hawkes models, culminating in Continuous Long Short Term Memory (cLSTM) recurrent neural networks.

About the speaker :

Conor Kresin is a lecturer of the Department of Mathematics and Statistics, University of Otago. His research interest include Point process theory and applications, stochastic geometry, disease modelling, information theory, causal inference.

Abstract:

Under model misspecification, it is known that Bayesian posteriors often do not properly quantify uncertainty about true or pseudo-true parameters. Even more fundamentally, misspecification leads to a lack of reproducibility in the sense that the same model will yield contradictory posteriors on independent data sets from the true distribution. To improve reproducibility, an easy-to-use and widely applicable approach is to apply bagging to the Bayesian posterior ("BayesBag"); that is, to use the average of posterior distributions conditioned on bootstrapped datasets. To define a criterion for reproducible uncertainty quantification under misspecification, we consider the probability that two confidence sets constructed from independent data sets have nonempty overlap, and we establish a lower bound on this overlap probability that holds for any valid confidence sets. We prove that credible sets from the standard posterior can strongly violate this bound, indicating that it is not internally coherent under misspecification, whereas the bagged posterior typically satisfies the bound. We demonstrate on simulated and real data.

About the speaker:

Jeff Miller is an Associate Professor of Biostatistics, Harvard University. He is interested in using statistics to understand the molecular mechanisms of diseases of aging. His methodological research focuses on robustness to model misspecification, nonparametric Bayesian models, frequentist analysis of Bayesian methods, and efficient algorithms for inference in complex models.

https://www.hsph.harvard.edu/profile/jeffrey-miller/

Abstract : I think the highest value skillset in statistical programming is knowing how to write good functions. Functions are often taught as a tool to avoid repetition using the mnemonic DRY: Don't Repeat Yourself. Whilst DRY is both true and real, I think functions are at their best when they encapsulate expression and are easy to reason with. That is, DRY is sufficient, but not necessary. Writing good functions is more than esoteric aesthetics. We need to be able to reason with our code in statistics. We often don't have the capacity to write tests to show our code is "correct". Instead, we need to rely on our ability to reason with, trust, and verify that the code works as it should. I believe writing good functions that encapsulate expressions and are able to be reasoned with are how we can ensure our code, and therefore our methods, and our analyses, work as they should. In this talk I will discuss some practical ideas on writing a good function, how to identify bad ones, and how to move between the two states.

About the speaker : I work as a research software engineer, with Nick Golding on the greta R package for statistical modelling, and implementing novel statistical methods for infectious diseases like COVID19 and malaria. I work at the Telethon Kids Institute, which is based in Perth, Western Australia, but I work remotely in Launceston, Tasmania. I am a strong advocate for free and open source software, and have written several R packages to improve data analysis.

https://www.njtierney.com/about/

Abstract: Dementia is a global health priority. The IDEA programme is a dementia prevalence study aiming to establish the true prevalence of dementia among older adults in Aotearoa New Zealand, with a particular emphasis on diverse ethnic groups. In this seminar, we will provide an overview of the methods and challenges associated with sampling older adults. We will discuss the various sampling strategies employed for each setting, including the community, retirement villages, and aged residential care.

Speaker: Ngaire Kerse is the Joyce Cook Chair in Ageing Well, a GP, and Professor of General Practice and Primary Health Care at the University of Auckland. With over 350 publications and 50 research grants, she is an international expert in falls prevention, bi-cultural ageing, and primary health care. Leading multiple research teams, Ngaire spearheads projects such as LiLACS NZ, focusing on equity, health service use, and well-being in advanced age. Her work on fall prevention includes studies on older individuals post-stroke and in residential care. Currently, she heads the IDEA programme, investigating the prevalence and impact of dementia in Aotearoa.

https://profiles.auckland.ac.nz/n-kerse

Abstract:

My term "regression averaging" refers to first running a regression estimation procedure, be it a linear model, k-Nearest Neighbors or whatever, then averaging the fitted values over some region. I will present two applications of this. The first is on the topic of dealing with missing values, specifically in a context of prediction rather than effect estimation. The second is in the area of removing bias with respect to sensitive variables, say race or gender in a prediction model.

About the speaker:

Norman Matloff is a professor in the Department of Computer Science at the University of California. Professor Matloff’s research areas include parallel processing (especially software distributed shared memory), statistical computing, and predictive analytics.

https://faculty.engineering.ucdavis.edu/matloff/

Abstract:

Current observations through ground-based detectors of gravitational waves (GWs) are having a pronounced effect on the understanding of our universe. Due to the presence of the earth, ground-based detectors are limited in sensitivity to lower frequency GWs, losing access to the rich science that can be reaped from higher mass black hole coalescences. The proposed space-based detector, the Laser Interferometer Space Antennae (LISA), eliminates sources of noise from the earth and will provide access to observations of GWs in the rich mHz frequency band, thus higher mass binaries. The aim of this talk is to be pedagogical in nature: reviewing GWs up to the first detection GW150914, providing an overview of LISA specific sources with a simple example of Bayesian inference applied to a toy GW model. We will finish on the prospects for the LISA instrument by discussing both current work and future challenges in the context of data analysis.

https://inspirehep.net/authors/1976434

Abstract:

The "predictmeans" R package provides a comprehensive set of functions for diagnosis and inference from a range of models common in statistical analysis. These models include those generated by "aov", "lm", "glm", "gls", "lme", "lmer", "glmer", "glmmTMB" and "semireg". Inferences include key statistical metrics such as predicted means and standard errors, contrasts, multiple comparisons, permutation tests, and adjusted R-squared values and graphical representations. This presentation will demonstrate the key capabilities of this package through practical examples, with a particular focus on semiparametric regression techniques and the calculation of adjusted R-squared values for generalized mixed-effects models.

https://www.researchgate.net/scientific-contributions/Dongwen-Luo-2004321262

Abstract :

Due to the advancement of computation power, sensor technologies, and data collection tools, the field of healthcare and public health monitoring and management have been evolved over the past several decades with different names under different application domains, such as statistical process control (SPC), process monitoring, health surveillance, prognostics and health management (PHM), personalized medicine, etc. There are tremendous opportunities in interdisciplinary research of system monitoring through integration of SPC, system informatics, data analytics, PHM, and personalized health management. In this talk we will present our views and experience in the evolution of systems monitoring and health management, its challenges and opportunities, as well as its applications in both healthcare surveillance and public health management.

About the speaker :

Kwok L Tsui is professor in the Grado Department of industrial and Systems Engineering at Virginia Polytechnic Institute and State University. Tsui’s current research interests include data science and data analytics, surveillance in healthcare and public health, personalized health monitoring, prognostics and systems health management, calibration and validation of computer models, process control and monitoring, and robust design and Taguchi methods.

https://www.ise.vt.edu/people/faculty/tsui.html

Abstract:

Gravitational waves (GWs) are ripples in space-time produced by some of the universe's most violent and energetic events. Since the first detection through ground-based detectors in 2015, they have opened a new pathway for understanding our universe. A new frequency range for detection will be opened by the Laser Interferometer Space Antenna (LISA), a space-based observatory expected to launch in the 2030s. One potential detection of LISA is the stochastic gravitational wave background (SGWB), which is the superposition of many unresolved GWs. Detecting SGWB holds significant promise in unravelling insights into the early universe and astrophysical sources. However, one of the most critical challenges is distinguishing it from LISA's stochastic instrumental noise. In order to solve this problem, we investigate the parameter estimation of SGWB and LISA noise using a Bayesian framework comprising parametric and non-parametric models. We propose three parametric models for SGWB: power law, broken power law and single peak. We fit the noise with a non-parametric model, using a prior based on a mixture of penalized splines (P-splines). They estimate spectral densities with sharp peaks and abrupt changes due to the flexibility of B-splines with knots placed based on the variation in the data. We combine a fixed number of B-splines with a simple difference penalty, which controls the degree of smoothness of power spectral density (PSD). We demonstrate accurate estimates of PSD in a simulation study and a case of realistic LISA noise, using only P-splines and will extend our analysis to full implementation of our model.

This is the PYR seminar.

Abstract:

This work explores enhancing healthcare for elderly patients in Aotearoa New Zealand through queueing theory and simulations, responding to the demographic shift towards an aging population. It addresses the need for more effective and equitable healthcare, considering workforce shortages and access disparities. By developing mathematical models for patient flow, waiting time, and resource allocation, this research underscores the necessity of models that adjust priorities and routing to alleviate service congestion, aiming to improve resource use, access equality, and elderly patient wellbeing.

This is Yvonne's PYR seminar.

Abstract :

Close-kin mark-recapture (CKMR) methods have recently been used to infer demographic parameters such as census population size and survival for fish of interest to fisheries and conservation. These methods have advantages over traditional mark-recapture methods as the mark is genetic, removing the need for physical marking and recapturing that may interfere with parameter estimation. For mosquitoes, the spatial distribution of close-kin pairs has been used to estimate mean dispersal distance, of relevance to vector-borne disease transmission and novel biocontrol strategies. Here, we extend CKMR methods to the life history of mosquitoes and comparable insects. We derive kinship probabilities for mother-offspring, father-offspring, full-sibling and half-sibling pairs, where an individual in each pair may be a larva, pupa or adult. A pseudo-likelihood approach is used to combine the marginal probabilities of all kinship pairs. To test the effectiveness of this approach at estimating mosquito demographic parameters, we develop an individual-based model of mosquito life history incorporating egg, larva, pupa and adult life stages. The simulation labels each individual with a unique identification number, enabling close-kin relationships to be inferred for sampled individuals. Using the dengue vector Aedes aegypti as a case study, we find the CKMR approach provides unbiased estimates of adult census population size, adult and larval mortality rates, and larval life stage duration for logistically feasible sampling schemes. Considering a simulated population of 3,000 adult mosquitoes, estimation of adult parameters is accurate when ca. 40 adult females are sampled biweekly over a three month period. Estimation of larval parameters is accurate when adult sampling is supplemented with ca. 120 larvae sampled biweekly over the same period. The methods are also effective at detecting intervention-induced increases in adult mortality and decreases in population size. As the cost of genome sequencing declines, CKMR holds great promise for characterizing the demography of mosquitoes and comparable insects of epidemiological and agricultural significance.

About the speaker :

John Marshall is a Professor in Residence of Biostatistics and Epidemiology whose research supports efforts to control and eliminate mosquito-borne diseases such as malaria, dengue, and Zika virus broadly.

https://publichealth.berkeley.edu/people/john-marshall/