Visual Analytics:An opportunity for the HPC community

Shawn J. Bohn

Shawn.Bohn@pnl.gov

September 8-10, 2008

HPC User Forum Meeting

Visual Analytics

Definitions, What and Why Visual Analytics

History of Science Leading into the Future

New Requirements within Digital Universe

Examples and Observations

Conclusion

2

3

Visual Analytics Definition

Congress: Visual analytics provides the last 12 inches between the masses of information and the human mind to make decisions

Science: Visual analytics is the science of analytical reasoning facilitated by interactive visual interfaces

4

History of Graphics and Visualization

70s to 80sCAD/CAM Manufacturing, cars, planes, and chips3D, education, animation, medicine, etc.

• 80s to 90s– Scientific visualization– Realism, entertainment

• 90s to 2000s– Information visualization– Web and Virtual environments

• 2000s to 2010s– Visual Analytics– Visual/audio appliances

5

Selected Societal Drivers and Observations

Scale of Things to Come: Information:

In 2002, recorded media and electronic information flows generated about 22 exabytes (1018) of informationIn 2006, the amount of digital information created, captured, and replicated was 161 EBIn 2010, the amount of information added annually to the digital universe will be about 988 EB (almost 1 ZB)

REFERENCES:A Forecast of Worldwide Information Growth Through 2010: IDCNational Open Source Enterprise - Intelligence Community Directive No. 301, July 11, 2006UC Berkeley School of Information Management and Systems: Now much Information

6

Why Must We Adapt

Scale of Things to Come: Information:

Drivers of Digital Universe:

70% of the Universe is being produced by individuals

Organizations (businesses, agencies, governments, universities) produce 30%

Wal-Mart has a database of 0.5 PB; it captures 30,000,000 transactions/day

The growth is unevenToday the United States accounts for 41% of the Universe; by 2010, the Asia Pacific region will be growing 40% faster than any of the other regions

7

Why Must We Adapt

Scale of Things to Come: Information

Drivers of Digital Universe

Kinds of Data:

About 2 GB of digital information is being produced per person per year

95% of the Digital Universe’s information is unstructured25% of the digital information produced by 2010 will be images

By 2010, the number of e-mailboxes will reach 2 billion The users will send 28 trillion e-mails/year, totaling about 6 EB of data

8

Why Must We Adapt

Scale of Things to Come: Information:

Drivers of Digital Universe:

Kinds of Data

Interaction:Today's interaction designed for point and click on individual items, directories, folders, and listsToday's interaction assumes user knows subject, concepts within information spaces, and can articulate what they wantToday's interaction assumes data and interconnecting relationships are static in meaning over timeToday's interaction is one way initiatedToday’s interaction (WIMP) designed over 30 years ago

Visual Analytic Engages Multiple Specialties

10

Examples Demonstrating Need

Changing Nature of Information Structure: Temporal, dynamically changing relationships, determination of intent (DC Sniper & ThemeRiver)

11

Examples Demonstrating Need

Information synthesis while preserving security and privacy Data signatures that are semantic and scale

Country AFirm 1

Firm 2Firm 3

Firm 4Firm 5Firm 6

Firm 7Firm 8

Firm 9Firm 10

A Bank

Financial

Images

Audio

Video

Discover what is there AND discover what isn’t there

HPC and Visual Analytics – Example

Krishnan M, SJ Bohn, WE Cowley, VL Crow, and J Nieplocha. 2007. "Scalable Visual Analytics of Massive Textual Datasets." In IPDPS 2007. IEEE International Parallel and Distributed Processing Symposium, 26-30 March 2007, Long Beach, CA, USA.

State-of-the-art1

Distributed in both processing and dataScalable (for a single data type)

LimitationsInability to reconfigure on the flyData model is based on subsettingBatch vs. interactive

12

Some Observations

End Users want to:Steer their data (INTERACTION)

Include and discard data on-the-fly

Via Scenarios/Hypothesis generation

Surround the user (i.e., data access)

Be data scale and modalities agnostic

Work from their desktops/mobile device

Web/thin client/server type applications

No specialized graphics hardware (simple visuals)

Customize visualization based on who they are.

Developers of Visual Analytics want:Better toolkits and APIs (e.g., Global Arrays) (who should do this?)

14

Conclusions

Visual Analytics is one of the fastest growing fields of study and practice. Practice of interdisciplinary science is requiredBroadly applies to many aspects of societyVisual Analytics is an HPC opportunity

Thank you