Uncertainty-Aware Data Transformations for Collaborative Reasoning Kwan-Liu Ma
Research Goal Develop mathematical foundations for uncertainty-aware data transformations to facilitate trustworthy collaborative reasoning using visual means
Proposed Tasks 1.Developing data transformation methods, coupled with uncertainty measures, for the extraction of relational and semantic structures of data 2.Modeling of uncertainty extraction, propagation and aggregation from transformations to reasoning 3.Studying applications for supporting Uncertainty-aware collaborative reasoning Mathematical modeling of uncertainty- aware visual analytics process
Evaluating Visual Analytics Process using Uncertainty Propagation Formalize the representation of uncertainty and basic operations Quantify, propagate, aggregate, and convey uncertainty through a series of data transformations Enhance and evaluate visual reasoning using uncertainty
An Uncertainty-Aware Evaluation Framework Sensitivity Modeling Data Sources Derived Data/ Abstractions Visual Elements Insight DATA/VISUAL TRANSFORMATIONS VISUAL MAPPING VIEW Sensitivity Coefficients Uncertainty Propagation Uncertainty on Derived Data Source Uncertainty Uncertainty Visual Mapping Uncertainty Views Sensitivity Analysis Comparing different transformation methods
Case 1: Geo-temporal Data Records on migration using boats over a period of 3 years Analysis to study landing patterns and estimate landing success rate Hypothesis: distance and time are correlated How much confidence can we place on our findings? Go-fast Rustic Raft
Data and Transformation Uncertainty Uncertainty: incomplete data, accuracy of distance computation Data Completion –Pair-wise deletion –Cluster based Distance estimation –Model dependent –Evaluation uses sensitivity analysis
Sensitivity Analysis
Case 2: Social Network Analytics Cell phone records for 400 people The goal is to characterize the sub-networks of interest linked to a particular cell phone user.
Structural Abstraction
Transformation using Relative Importance
Uncertainty Analysis of the Importance Transform Node size: importance Halos and error bars: uncertainty Pie chart: sensitivity parameters wrt connected nodes Weighted based on number of calls 200 mostly depends on 5, 137 and 2
Impact of the Proposed Work First work addressing uncertainty throughout the whole visual analytics process Providing a more trustworthy view of the data A framework for cross comparison of different data transformation methods Providing a mechanism for assessing “what if” scenarios
Plans for Developing FODAVA Research collaborations with FODAVA lead institute and RVAC centers Demonstration and dissemination of research results through a variety of mechanisms Other outreach and education activities –Industry –VisWeek/VAST –SIGGRAPH, KDD, ICDM, …
Kwan-Liu Ma