Please join us online or in person for the November Queensland Branch Meeting. Details for the seminar are provided below. After the talk, come catch up with us in person for dinner and an early Christmas celebration.
TITLE: Test anything with Universal Inference
SPEAKER: Dr Hien Nguyen, University of Queensland
TIME: 4:00 pm - 5:00 pm (AEST), Wednesday 2nd November 2022
VENUE: Q430, Q block, Kelvin Grove Campus, Queensland University of Technology or Online (Zoom details will be sent with registration)
Please note that the seminar will be recorded and might be put on YouTube or similar platform.
ABSTRACT:
Two of the most important tasks in statistical inference are the testing of null hypotheses and the construction of confidence sets for distributional parameters and properties of interest. In simple scenarios, these tests and confidence sets may be constructed from elementary probability statements, but for many practical applications, tests and confidence sets rely on the availability of asymptotic theorems that may be difficult to justify in some situations and may not yield useful devices in others. The latter outcome is typical when the properties or parameters of interest take on values on the boundaries of parameter spaces or are contained in atypical manifolds of the Euclidean space.
The universal inference approach has recently been proposed as a means of constructing finite sample valid hypothesis tests and confidence sets when typical asymptotics do provide feasible inferential constructions. The approach hinges on the construction of an "E-value", using a clever hold-out argument, not dissimilar to the popular cross validation approach for model selection. We shall demonstrate the usefulness of the universal inference approach in a variety of inferential settings, including for time-series, composite likelihoods, and empirical Bayes.
SPEAKER'S BIO:
Hien Nguyen is a Senior Lecturer at the University of Queensland, and previously a Senior Lecturer at La Trobe University in Melbourne, where he was also an SSA Vic Council member in 2021. Hien was previously an ARC DECRA Fellow from 2017 to 2020 and is currently serving as the Statistical Computing and Technical Editor for the society's journal: The Australian and New Zealand Journal of Statistics. Hien's research focuses on optimization algorithms for complex and heterogeneous statistical and machine learning models, as well as inferential tools for such models, especially in the cases when traditional asymptotic techniques fail. He has applied his statistical expertise to produce novel and useful methods in neuroimaging, agriculture, genetics, proteomics, economics, and public health research. When not teaching or researching, Hien most enjoys spending time with his two cats: Nyx and Penelope.
@StatSocAus
