Ursula Laa- Research Fellow

Hi, I am Ursula, a researcher in the Statistics Institute of the University of Natural Resources and Life Sciences (BOKU) in Vienna. I focus on new methods and applications in data science with emphasis on interdisciplinary work, for example connecting my background in physics and my current interest in statistical data visualisation.
I mostly work on high-dimensional problems and use methods like the grand tour for visualisation of data beyond 3D. We can gain new physics insights by using better visualisations, but at the same time thinking about how to best plot the result also motivates us to develop new visualisation methods! Have a look through my recent publications for some interesting examples.
As part of my research I use and develop open source software, mostly in R or python. I am currently developing and mainitaining several R packages. A selection is given below, but you can check out my GitHub account to get a full overview.
Main interests: Data visualisation, statistical graphics, high-dimensional data analysis, machine learning and (particle) physics

Recent Publications

This is only a selection of some of my recent papers. You can find a complete list in my CV.

A Review of the State-of-the-Art on Tours for Dynamic Visualization of High-dimensional Data, WIREs Computational Statistics. [code] Stuart Lee, Dianne Cook, Natalia Da Silva, Ursula Laa, Earo Wang, Nick Spyrison, H Sherry Zhang A review of tour methods including the theory, history, recent developments and applications.
Pandemonium: a clustering tool to partition parameter space--application to the B anomalies, EPJ Plus. [code] Ursula Laa, German Valencia Clustering can be used to understand the interplay between observable and parameter space and we use a set of numeric and visual summaries to interactively explore this for examples in particle physics phenomenology.
Burning sage: Reversing the curse of dimensionality in the visualization of high-dimensional data, JCGS. [code] Ursula Laa, Dianne Cook, Stuart Lee When we visualize high-dimensional data with low-dimensional projections, we typically see piling of observations near the center. This work introduces a radial transformation that redistributes the points in the projection plane to reverse these piling effects.
Hole or grain? A Section Pursuit Index for Finding Hidden Structure in Multiple Dimensions, JCGS. [code] Ursula Laa, Dianne Cook, Andreas Buja, German Valencia We commonly visualize multivariate data with linear projections, but they can obscure features like high or low density regions near the center of a distribution. In this article we describe section pursuit, a method that finds slices revealing these hidden structures.
A slice tour for finding hollowness in high-dimensional data, JCGS. [code] [animations] Ursula Laa, Dianne Cook, German Valencia Sectioning, or slicing, through high dimensions is complementary to projection based visualisation methods, and can reveal otherwise hidden features, for example in concave distributions. This note describes a new approach to slicing in the orthogonal space of 2D projections while running a tour. Example animations show slices through hollow geometric objects and how features hidden at the center of a distribution can be found with slicing.
Using tours to visually investigate properties of new projection pursuit indexes with application to problems in physics, Computational Statistics. [code] Ursula Laa, Dianne Cook In this paper we develop a framework for evaluating the performance of new projection pursuit index functions and we apply it to interesting new indexes that detect complex bivariate patterns in projections. The potential of these index functions is demonstrated with examples from gravitational wave astronomy.
Connecting R with D3 for dynamic graphics, to explore multivariate data with tours, The R Journal. [code] Michael Kipp, Ursula Laa, Dianne Cook In this paper we explore using custom messages to pass data from R to D3, using the Shiny framework. This is an approach can generally be used for creating interactive graphics, and we use it for a new GUI for the tourr package.
Anatomy of a six-parameter fit to the b → sl+l− anomalies, EPJC. [animations] Bernat Capdevila, Ursula Laa, German Valencia Measurements of processes with an underlying quark transition b → sl+l− show some of the most interesting deviations from the Standard Model of particle physics. In this paper we show how we can understand the measurements in the context of a 6D global fit, and accounting for correlation effects. The results provide guidance for future theoretical and experimental work.
Dynamical projections for the visualisation of PDFSense data, EPJC. [animations] Dianne Cook, Ursula Laa, German Valencia This paper presents the first application of tour methods to a problem in theoretical particle physics. We use dynamic visualisations to understand the impact of single measurements on a global fit of parton distribution functions, for example demonstrating grouping and highlighting outliers. These animations in gif format can be downloaded from the auxiliary material on arXiv.

Software

This is a selection of the R packages that I am maintaining or have made important contributions to.

tourr [code] The tourr package implements tour algorithms and displays and I have contributed new displays and guidance methods.
pandemonium [code] A Shiny app for the interactive exploration of hierarchical clustering results with different settings.
binostics [code] The binostics package calculates graph theoretic scagnostics measures. It was written by Hadley Wickham and I am currently maintaining the package.
galahr [code] A GUI for the tourr package in R, using shiny and plotly. This package makes it easy for new users to work with tours for the visualisation of high-dimensional data, and also introduces useful interactive features. (Note that the app needs to be updated to work well with recent versions of plotly.)
spinebil [code] spinebil provides functionalities to evaluate the performance of projection pursuit index functions using tour methods.

Recent Talks

Slides for some of my recent talks are listed below. For slides made with R markdown the source code is also linked.

Section pursuit, presented at CMStatistics 21. [code]
Tours for the dynamic visualization of high-dimensional data, presented at Rencontres R 2021. [code]
New displays for the visualization of multivariate data in the tourr package, presented at UseR! 2021. [code]
Burning sage: Reversing the curse of dimensionality in the visualization of high-dimensional data, presented at the 2020 ACEMS virtual retreat. [code]
High-dimensional data visualisation with tours, presented at the Data Visualisation New York meetup. [code]
High-dimensional data visualisation with the grand tour, presented at CHEP 2019. [code]
An Introduction to the Visualisation Ecosystem in R, ACEMS technical seminar with Stuart Lee. [code]
Visualising high-dimensional data - New developments of the tourr package using Shiny and plotly, presented at UseR! 2019. [code]
Visualising physics - High-dimensional data visualisation for physics applications, seminar talk. [code]