Stanford Seminar - Theories of Inference for Visual Analysis

Explore theories of inference for visual analysis in this Stanford seminar. Delve into how visualizations offload cognition to perception and examine state-of-the-art approaches for optimizing pattern finding. Investigate methods for visualizing distribution and uncertainty, including frequency-based techniques. Evaluate effect size judgments and the mean difference heuristic. Analyze Bayesian cognition in belief updating and examine how visualizations support causal inference and learning in analysis. Discover how good visualizations facilitate model checks and explore the concept of graphical inference as a Bayesian model check. Learn about a Tableau-like system supporting model checks and the process of deriving implied models from visualizations. Consider open questions in the field, including characteristics of good inference problems for visualization and methods for bounding the value of more expressive visualizations.

Introduction.
Visualizations help offload cognition to perception.
State-of-the-art: Optimize for easy pattern finding.
Approaches to visualize distribution or uncertainty.
Error bars/intervals violate expressiveness.
Probability encodings often hard to read accurately.
Frequency based uncertainty visualization.
Evaluation: Effect size judgments.
Mean difference heuristic in effect size judgments.
Sensitivity to visual distance between means.
Investigating the mean difference heuristic.
Uncertainty visualization conditions.
Bayesian cognition to understand belief updating.
How well do visualizations support causal inference?.
How does visualization support learning in analysis?.
Good visualizations facilitate model checks.
Implicit model checks give graphs their meaning.
Graphical inference as Bayesian model check.
Tableau-like system supporting model checks.
From visualization to implied model.
Open questions.
What characterizes good inference problems for vis?.
Bounding the value of a more expressive visualization.