TITLE:
Interactive Data Analysis

ABSTRACT:
In this talk, I will discuss our research on interactive systems for 
data cleaning, assessment and exploration. Collectively, these systems
contribute new approaches for improving the efficiency and scale at 
which expert analysts work, while lowering the threshold for 
non-experts.

Data analysts often expend an inordinate amount of effort manipulating
data and assessing data quality issues. With our Wrangler system, users
construct data transformation scripts in a direct manipulation 
interface. Wrangler uses programming-by-demonstration methods to 
automatically suggest applicable transforms and preview their output. 
The end result is not simply transformed data, but a reusable program 
that can be run at scale. Once data has been suitably transformed, our
Profiler system combines anomaly detection and automatic visualization
recommendations to aid quality assessment.

Next, I will describe imMens, a system for real-time visual querying of
big data. To enable visual analysis at scale, imMens follows the 
principle that perceptual and interactive scalability should be limited
by the chosen resolution of the visualized data, not the number of
records. Using a combination of multivariate data tiles and client-side
parallel query processing, imMens supports 50 frames-per-second linked 
selection across summary visualizations of data sets ranging from
thousands to billions of records.