-
My Model is Unfair, Do People Even Care? Visual Design Affects Trust and Perceived Bias in Machine Learning
Authors:
Aimen Gaba,
Zhanna Kaufman,
Jason Chueng,
Marie Shvakel,
Kyle Wm. Hall,
Yuriy Brun,
Cindy Xiong Bearfield
Abstract:
Machine learning technology has become ubiquitous, but, unfortunately, often exhibits bias. As a consequence, disparate stakeholders need to interact with and make informed decisions about using machine learning models in everyday systems. Visualization technology can support stakeholders in understanding and evaluating trade-offs between, for example, accuracy and fairness of models. This paper a…
▽ More
Machine learning technology has become ubiquitous, but, unfortunately, often exhibits bias. As a consequence, disparate stakeholders need to interact with and make informed decisions about using machine learning models in everyday systems. Visualization technology can support stakeholders in understanding and evaluating trade-offs between, for example, accuracy and fairness of models. This paper aims to empirically answer "Can visualization design choices affect a stakeholder's perception of model bias, trust in a model, and willingness to adopt a model?" Through a series of controlled, crowd-sourced experiments with more than 1,500 participants, we identify a set of strategies people follow in deciding which models to trust. Our results show that men and women prioritize fairness and performance differently and that visual design choices significantly affect that prioritization. For example, women trust fairer models more often than men do, participants value fairness more when it is explained using text than as a bar chart, and being explicitly told a model is biased has a bigger impact than showing past biased performance. We test the generalizability of our results by comparing the effect of multiple textual and visual design choices and offer potential explanations of the cognitive mechanisms behind the difference in fairness perception and trust. Our research guides design considerations to support future work developing visualization systems for machine learning.
△ Less
Submitted 7 August, 2023;
originally announced August 2023.
-
Professional Differences: A Comparative Study of Visualization Task Performance and Spatial Ability Across Disciplines
Authors:
Kyle Wm. Hall,
Anthony Kouroupis,
Anastasia Bezerianos,
Danielle Albers Szafir,
Christopher Collins
Abstract:
Problem-driven visualization work is rooted in deeply understanding the data, actors, processes, and workflows of a target domain. However, an individual's personality traits and cognitive abilities may also influence visualization use. Diverse user needs and abilities raise natural questions for specificity in visualization design: Could individuals from different domains exhibit performance diff…
▽ More
Problem-driven visualization work is rooted in deeply understanding the data, actors, processes, and workflows of a target domain. However, an individual's personality traits and cognitive abilities may also influence visualization use. Diverse user needs and abilities raise natural questions for specificity in visualization design: Could individuals from different domains exhibit performance differences when using visualizations? Are any systematic variations related to their cognitive abilities? This study bridges domain-specific perspectives on visualization design with those provided by cognition and perception. We measure variations in visualization task performance across chemistry, computer science, and education, and relate these differences to variations in spatial ability. We conducted an online study with over 60 domain experts consisting of tasks related to pie charts, isocontour plots, and 3D scatterplots, and grounded by a well-documented spatial ability test. Task performance (correctness) varied with profession across more complex visualizations, but not pie charts, a comparatively common visualization. We found that correctness correlates with spatial ability, and the professions differ in terms of spatial ability. These results indicate that domains differ not only in the specifics of their data and tasks, but also in terms of how effectively their constituent members engage with visualizations and their cognitive traits. Analyzing participants' confidence and strategy comments suggests that focusing on performance neglects important nuances, such as differing approaches to engage with even common visualizations and potential skill transference. Our findings offer a fresh perspective on discipline-specific visualization with recommendations to help guide visualization design that celebrates the uniqueness of the disciplines and individuals we seek to serve.
△ Less
Submitted 4 August, 2021;
originally announced August 2021.
-
Property Decoupling across the Nucleus-Melt Interface during Polymer Crystal Nucleation
Authors:
Kyle Wm. Hall,
Simona Percec,
Wataru Shinoda,
Michael L. Klein
Abstract:
Spatial distributions are presented that quantitatively capture how polymer properties (e.g., segment alignment, density, and potential energy) vary with distance from nascent polymer crystals (nuclei) in prototypical polyethylene melts. It is revealed that the spatial extent of nuclei and their interfaces is metric-dependent as is the extent to which nucleus interiors are solid-like. As distance…
▽ More
Spatial distributions are presented that quantitatively capture how polymer properties (e.g., segment alignment, density, and potential energy) vary with distance from nascent polymer crystals (nuclei) in prototypical polyethylene melts. It is revealed that the spatial extent of nuclei and their interfaces is metric-dependent as is the extent to which nucleus interiors are solid-like. As distance from a nucleus increases, some properties, such as density, decay to melt-like behavior more rapidly than polymer segment alignment, indicating that a polymer nucleus resides in a nematic-like droplet. This nematic-like droplet region coincides with enhanced formation of ordered polymer segments that are not part of the nucleus. It is more favourable to find non-constituent ordered polymer segments near a nucleus than in the surrounding metastable melt, pointing to the possibility of one nucleus inducing the formation of other nuclei. These findings provide a conceptual bridge between polymer crystal nucleation under non-flow and flow conditions, and are used to rationalize previous results.
△ Less
Submitted 4 March, 2020;
originally announced March 2020.
-
Divining the Shape of Nascent Polymer Crystal Nuclei
Authors:
Kyle Wm. Hall,
Timothy W. Sirk,
Simona Percec,
Michael L. Klein,
Wataru Shinoda
Abstract:
We demonstrate that nascent polymer crystals (i.e., nuclei) are anisotropic entities, with neither spherical nor cylindrical geometry, in contrast to previous assumptions. In fact, cylindrical, spherical, and other high symmetry geometries are thermodynamically unfavorable. Moreover, post-critical transitions are necessary to achieve the lamellae that ultimately arise during the crystallization of…
▽ More
We demonstrate that nascent polymer crystals (i.e., nuclei) are anisotropic entities, with neither spherical nor cylindrical geometry, in contrast to previous assumptions. In fact, cylindrical, spherical, and other high symmetry geometries are thermodynamically unfavorable. Moreover, post-critical transitions are necessary to achieve the lamellae that ultimately arise during the crystallization of semicrystalline polymers. We also highlight how inaccurate treatments of polymer nucleation can lead to substantial errors (e.g., orders of magnitude discrepancies in predicted nucleation rates). These insights are based on quantitative analysis of over four million crystal clusters from the crystallization of prototypical entangled polyethylene melts. New comprehensive bottom-up models are needed to capture polymer nucleation.
△ Less
Submitted 5 August, 2019;
originally announced August 2019.
-
Design by Immersion: A Transdisciplinary Approach to Problem-Driven Visualizations
Authors:
Kyle Wm. Hall,
Adam J. Bradley,
Uta Hinrichs,
Samuel Huron,
Jo Wood,
Christopher Collins,
Sheelagh Carpendale
Abstract:
While previous work exists on how to conduct and disseminate insights from problem-driven visualization projects and design studies, the literature does not address how to accomplish these goals in transdisciplinary teams in ways that advance all disciplines involved. In this paper we introduce and define a new methodological paradigm we call design by immersion, which provides an alternative pers…
▽ More
While previous work exists on how to conduct and disseminate insights from problem-driven visualization projects and design studies, the literature does not address how to accomplish these goals in transdisciplinary teams in ways that advance all disciplines involved. In this paper we introduce and define a new methodological paradigm we call design by immersion, which provides an alternative perspective on problem-driven visualization work. Design by immersion embeds transdisciplinary experiences at the center of the visualization process by having visualization researchers participate in the work of the target domain (or domain experts participate in visualization research). Based on our own combined experiences of working on cross-disciplinary, problem-driven visualization projects, we present six case studies that expose the opportunities that design by immersion enables, including (1) exploring new domain-inspired visualization design spaces, (2) enriching domain understanding through personal experiences, and (3) building strong transdisciplinary relationships. Furthermore, we illustrate how the process of design by immersion opens up a diverse set of design activities that can be combined in different ways depending on the type of collaboration, project, and goals. Finally, we discuss the challenges and potential pitfalls of design by immersion.
△ Less
Submitted 17 October, 2019; v1 submitted 1 August, 2019;
originally announced August 2019.
