A Mathematical Framework and Software Implementation for Uncertainty Visualisation
summary
The gist
Random variables are the bread and butter of statistics, and visualisations are one of the most versatile tools in the field…
In short
The authors developed a mathematical framework to visualize random variables by treating visualizations as continuous functions, integrating uncertainty into the grammar of graphics. They implemented this theory in the R package 'ggdibbler', which allows users to replace data with random variables while guaranteeing statistical convergence properties.
Key concepts
- Continuous Mapping Theorem (CMT)
- This theorem ensures that visualisations should behave similarly to their underlying distributions. The framework requires visual functions to obey the CMT, meaning the way a plot changes when the input data changes should be mathematically predictable and consistent with statistical theory.
- Visual Function
- The visual function is a composite mathematical formula that maps an input matrix (data) from the sample space to a set of visual statistics. It is decomposed into steps like scaling, statistics, geometry conversion, and coordinate transformation to handle both deterministic and random inputs.
- Visual Convergence
- This concept defines when two random graphics are considered visually converged—meaning their rendered outputs are indistinguishable according to a human perception metric. The framework uses this to automatically manage sample sizes so that conclusions remain stable across different renderings of the plot.
Terminology used across episodes
This episode discusses
- A Mathematical Framework and Software Implementation for Uncertainty Visualisation · Paper Radio
- The Noisy Work of Uncertainty Visualisation Research: A Review · Paper Radio
The paper
A Mathematical Framework and Software Implementation for Uncertainty Visualisation · Read on arXiv
Monash University
Random variables are the bread and butter of statistics, and visualisations are one of the most versatile tools in the field, so it is a wonder why we do not have a methodology for visualising random variables. This gap is particularly evident for exploratory data analysis (EDA). We address this gap by designing a mathematical framework for visualisation, which argues that we should consider visualisations to be continuous functions. In the case of random variable inputs, this means the visualisations should obey the continuous mapping theorem (CMT). By breaking the visual function down into its components, we are able to identify which parts of the mapping are ill-defined for random variable inputs and redefine them in a way that guarantees both the flexibility required for EDA and the statistical sensibility of CMT. This formalisation represents a complete integration of uncertainty into the grammar of graphics, which we show by implementing the theory in the R package, "ggdibbler". The ggdibbler software is a "ggplot2" extension that allows users to replace the data of any plotting function with a random variable, with the guarantee that the visualisation will have the same convergence properties as its underlying data.
DOI: 10.1080/10618600.2026.2759709
Transcript
Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.
Tom: We're looking at the paper: "A Mathematical Framework and Software Implementation for Uncertainty Visualisation".
Jane: Here's the gist of the paper. Random variables are the bread and butter of statistics, and visualisations are one of the most versatile tools in the field, so it is a wonder why we do not have a methodology for visualising random variables The authors address this gap by designing a mathematical framework for visualisation, which argues that we should consider visualisations to be continuous functions This formalisation represents a complete integration of uncertainty into the grammar of graphics, which they show by implementing the theory in the R package, “ggdibbler” Pruning redundant reasoning tokens The paper introduces a mathematical framework for visualising random variables based on the continuous mapping theorem (CMT) to ensure statistical sensibility in graphics The core idea is that visualisations should obey the CMT, meaning they should have similar convergence properties to their underlying distributions This framework breaks down the visual function into components to identify parts ill-defined for random variable inputs and redefine them for flexibility and statistical sensibility The implementation is provided in the R package “ggdibbler”, which allows users to replace data with a random variable while guaranteeing convergence properties How the training signal is built The visual function is formally defined as a composite function, decomposing it into steps like scaling, statistics, geometry conversion, coordinate system transformation, and an injective function This deterministic visual function maps a deterministic matrix to a statistical graphic For uncertainty visualisation, the input is a random matrix where distributions replace single values The framework requires adjustments to scales, statistics, and geometry to generalize the deterministic visual function for random matrix inputs while upholding Theorem 3.1 Where the gains come from The paper details several necessary adjustments to generalize the grammar of graphics for uncertainty visualisation 1.
Tom: The adjustment to scales involves a "change of variable" where the scale function M is applied to X, resulting in M(X) with the induced probability measure PM(X)(A) = P (M−1(A)) 2.
Jane: The adjustment to statistics requires defining two functions: one representing the distribution and one summarizing the data, defined as S = Ssample ∘ S∗ 3.
Tom: The adjustment to geometry involves a generalized geometry function G∗ that transforms an array of data into a matrix of geometric positions, decomposed into within sample and between sample position adjustments Pruning redundant reasoning tokens The software design leverages object-oriented programming (OOP) principles to handle the polymorphism required for uncertainty visualisation, ensuring the visual function is consistent regardless of input type The implementation in ggdibbler uses a suffix like “ sample” to maintain similar naming conventions with ggplot2 while explicitly indicating that only the input has changed The system is designed such that users should have quantified uncertainty as distributions before attempting visualisation, leveraging the distributional package to store distributions inside data frames as individual cells Where the gains come from The paper concludes by outlining future research directions, suggesting investigation into bias/variance trade-off and determining the minimum level of variance required for human perception of visual convergence It also questions whether a more fundamental version of the grammar of graphics exists that can make the link between data and visual aesthetic more explicit The software has room for improvement by expanding to accept joint distributions and integrating with ggplot2 extensions Pruning redundant reasoning tokens The paper formalises uncertainty visualisation by defining a generalized visual function that accepts random variable inputs, which is implemented in the ggdibbler R package This framework ensures that plots always obey Theorem 3.1, maintaining the connection between data and visual aesthetic The system manages computational complexity by controlling sample size via the times argument, advising users to select a size large enough for visual convergence The emergent aesthetics, such as blur and fuzziness in ggdibbler plots, result from the combination of transparency and randomness rather than direct mapping of uncertainty to an aesthetic The paper emphasizes that both ggdist and ggdibbler are required for a complete picture of uncertain data because they represent orthogonal approaches to signal and noise visualization The full list of planned changes for the package is available in the GitHub repository The paper acknowledges that direct mapping uncertainty to an aesthetic might be computationally effective but is fraught with danger and results in more headaches than it would be worth The first author's work was supported by a scholarship from the Australian Energy Market Operator The research was supported by the Commonwealth through an Australian Government Research Training Program Scholarship
DOI: https://doi.org/10.82133/C42F-K220: The R packages used for this work included tidyverse, distributional, ggdist, ggdibbler, and others The GitHub repository for this paper can be found at https://github.com/harrietmason/paper-ggdibbler The paper was written on June 23, 2026.
Title and authors: Tom: So this paper is called A Mathematical Framework and Software Implementation for Uncertainty Visualisation, and it looks like they are trying to build a proper method for visualizing random variables.
Jane: That's right, and the core idea they're pushing is that visualizations should be thought of as continuous functions. They argue that this approach means we need to follow something called the continuous mapping theorem, or CMT, when dealing with random variable inputs.
Lu: The authors are breaking down the visual function into its different parts so they can spot where things aren't clearly defined for random variables and redefine those parts to make it flexible and statistically sound.
Meng: So what does that actually mean for us when we're just trying to make a plot? Does it give us a new tool or just more complicated theory?
Tom: It’s giving them a mathematical structure, essentially treating the visualization process itself as a continuous function, which should give us more control over how we handle uncertainty in our graphics.
Jane: Exactly. They want to integrate uncertainty directly into the grammar of graphics, and they show this by implementing their theory in an R package called ggdibbler.
The paper's summary: Tom: So what they're saying here is that instead of just making more charts, the real issue with uncertainty visualization isn't just having more plot names or software, but the structure and flexibility of the system itself.
Jane: They’re saying that current implementations are brittle and constrained by their own set of assumptions, which limits how we can explore uncertain data from different sources or dimensions.
Lu: They suggest a useful system needs to let you explore with uncertain data from multiple different sources, dimensions, and data types, which is a flexibility that all the current implementations fail to provide.
Meng: So it’s not about inventing a new plot type; it’s about building a flexible framework that can handle more complex data inputs without breaking.
Tom: Right. And they address this by assuming the input for uncertainty visualization is just a dataset where every single cell is treated as a random variable, containing all the quantified uncertainty we want to represent.
Jane: And they make a distinction between two ways of looking at things, which is signal and noise: one approach, like ggdist plot, looks at the shape of the distribution of each observation to visualize it as a signal.
Lu: Whereas the ggdibbler plot focuses on how that uncertainty would change the conclusions from a standard ggplot2 plot, visualizing it instead as noise.
Meng: So they are presenting two orthogonal approaches to visualization that aren't just variations on the same theme.
The paper's improvements: Tom: Now they detail several specific adjustments needed to generalize the grammar of graphics for this uncertainty visualization, and those adjustments include changing how scales are handled.
Jane: They do this by applying a change of variable, where they use a scale function M on the data X to get M of X, which induces a probability measure PM on A as the probability P of that new matrix being less than or equal to some value.
Lu: Then for statistics, they define two functions: one that represents the actual distribution and another that summarizes the data into a statistical summary, defined as S is sample composed with S star.
Meng: And geometry gets tricky too; they introduce a generalized geometry function G star that transforms an array of data into a matrix of geometric positions by adjusting both within sample and between sample positions.
Tom: So these adjustments are necessary to generalize the deterministic visual function, which maps a deterministic matrix to a statistical graphic, for these random inputs while still upholding Theorem three point one.
Jane: And they use object-oriented programming principles in the software design to handle the polymorphism needed for uncertainty visualization, making sure the visual function stays consistent regardless of what kind of input you give it.
Conclusion: Tom: So, to wrap up, this paper formalizes uncertainty visualization by defining a generalized visual function that accepts random variable inputs and is implemented in the ggdibbler R package.
Jane: This framework ensures that all plots always obey Theorem three point one, which maintains the connection between the data and the visual aesthetic they’re trying to achieve.
Lu: The system manages computational complexity by controlling sample size through a times argument, suggesting users pick a size large enough for visual convergence to ensure conclusions don't change between renderings of the plot.
Meng: And they note that the blur and fuzziness you see in ggdibbler plots is more of an emergent aesthetic coming from transparency and randomness rather than just directly mapping uncertainty to an aesthetic.
Tom: They also point out that both ggdist and ggdibbler are necessary for a complete picture because they represent orthogonal approaches to signal and noise visualization.
Jane: The paper concludes by suggesting future research into the bias variance trade-off, specifically determining the minimum level of variance needed for human perception of visual convergence.
Lu: They also question whether there's a more fundamental version of the grammar of graphics that could make the link between data and visual aesthetic more explicit.
Meng: From an engineering standpoint, they admit that direct mapping uncertainty to an aesthetic might be computationally effective, but it’s fraught with danger and just ends up being a lot of extra work.
Tom: So, this paper is about building a rigorous mathematical foundation for how we visualize random variables using the ggdibbler R package.
More episodes
- Daily Summary for 2026-11-19
- 2310.07973-Statistical Performance Guarantee for Subgroup Identification with Generic Machine Learning
- 2408.09155-Learning Robust Treatment Rules for Censored Data
- Daily Summary for 2026-11-19
- 2603.23397-Kinetic Langevin Splitting Schemes for Constrained Sampling
- 2606.09906-An information-geometric framework for mapping maximum potential biodiversity
- 2603.27718-Induced replication and the assessment of models
- 1307.0898-Entropy of a Zipfian Distributed Lexicon
- 2406.19974-Generalizing self-normalized importance sampling with couplings
- 2412.10213-Collaborative Optimal Design of Controlled Experiments with Subject Covariates