InfoVis at UBC CS

1
InfoVis at UBC CS

• Tamara Munzner
• current graduate students
• Dan Archambault
• Aaron Barsky
• Stephen Ingram
• Heidi Lam
• Peter McLachlan
• James Slack

2
Research Agenda

• problem information explosion
• sensors, logging, simulation, business data...
• solution information visualization (infovis)
• help people accomplish tasks more effectively by
  exploiting human perceptual system to aid
  cognition through the graphic representation of
  abstract data
• display relevant information graphically to
  assist in memory tasks
• support data exploration through direct
  interaction
• assist in pattern finding through the display of
  overview and detail, search, and user-directed
  reordering
• major theme in my work scalability

3
Research Philosophy

• emphasis on collaboration
• people with driving problems
• opportunistic, find people with big data and
  clear questions
• past topology, linguistics, web site design,
  environmental sustainability, evolutionary
  biology
• current bioinformatics, networking, computer
  systems, data analysis
• psychologists, HCI people for evaluation
• other infovis or graphics people
• often release software so informed by user
  community needs
• usually open-source, sometimes proprietary

4
Current Research

• accordion drawing
• started with evolutionary trees, gene sequences
• monitoring large collections of machines
• log visualization
• analyzing large session logs
• dimensionality reduction
• finding high-dimensional clusters quickly using
  GPU
• network visualization
• general multi-level graph drawing
• specialized protein-protein interaction networks
• infovis evaluation
• understanding when and how these techniques work
  as planned

5
Accordion Drawing

• stretch-and-squish navigation
• rubber sheet with borders nailed down
• Sarkar et al 93, ...
• integrate overview, details
• guaranteed visibility
• marks always visible
• important for scalability
• new idea
• Munzner et al 03

6
Guaranteed Visibility Challenges

• easy for small datasets
• hard with larger ones
• reasons a mark could be invisible
• outside the window
• AD solution constrained navigation
• underneath other marks
• AD solution avoid 3D
• smaller than a pixel
• AD solution smart culling

7
Accordion Drawing Applications

• TreeJuxtaposer
• side by side visual comparison of evolutionary
  trees
• SequenceJuxtaposer
• multiple aligned gene sequences

downloadable from olduvai.sourceforge.net
joint work with James Slack, Kristian Hildebrand,
Katherine St. John (CUNY)
8
Accordion Drawing Applications

• LiveRAC
• monitoring huge computer clusters

joint work with Peter McLachlan, Stephen North
(ATT), Elefterios Koutsofios (ATT)
9
LiveRAC Problem Domain

• Managed hosting services, network operations
  centre staff

10
Monitored Data

• Most data collected from monitored network
  devices is time-series data
• any type of computer or appliance servers,
  routers
• time stamp and value
• Two types of time-series objects collected
• performance metrics
• 10 AUG 2006 95237, CPU, 95
• alarm data
• 10 AUG 2006 95237, MAJOR, HIGH TEMP
• Key difference for visualization
• performance metrics quantitative
• alarms categorical

11
Visualization Solution Requirements

• Scale to large, dynamic time-series datasets
• thousands of devices
• dozens of data channels
• Interact with previously gathered data

Active region Time scale of items
Total database Days to years Billions
In memory Several seconds Millions
On screen Sub-second Thousands
DB (SWIFT)
LiveRAC
12
Our Solution LiveRAC

• Interactive user-directed exploration of
  overview detail
• rapidly explore time-series data with context
  available at all times
• live demo

13
Semantic Zooming and Aggregation

• compact representations in reduced areas
• large cells show time-series charts
• aggregate spatial representation shown in highly
  compressed regions

14
LiveRAC Demo
15
Current Research

• accordion drawing
• started with evolutionary trees, gene sequences
• monitoring large collections of machines
• log visualization
• analyzing large session logs
• dimensionality reduction
• finding high-dimensional clusters quickly using
  GPU
• network visualization
• general multi-level graph drawing
• specialized protein-protein interaction networks
• infovis evaluation
• understanding when and how these techniques work
  as planned

16
Session Viewer Log Visualization
metadata
session logs
data
joint work with Heidi Lam, Diane Tang (Google),
Dan Russell (Google)
17
Session Viewer Log Visualization

• What To develop a visualization tool to analyze
  session data
• Why Session data analysis is hard because of the
  volume and complexity of the data. Statistics
  does not tell the whole story, but detailed
  session-by-session analysis is impossible.
• How Harness human visual capabilities to spot
  potentially interesting trends/patterns, to allow
  analysts to focus on a manageable subset of the
  data in detail

18
Session Viewer Video
19
Current Research

• accordion drawing
• started with evolutionary trees, gene sequences
• monitoring large collections of machines
• log visualization
• analyzing large session logs
• dimensionality reduction
• finding high-dimensional clusters quickly using
  GPU
• network visualization
• general multi-level graph drawing
• specialized protein-protein interaction networks
• infovis evaluation
• understanding when and how these techniques work
  as planned

20
Dimensionality Reduction

• mapping high-dimensional space into space of
  fewer dimensions
• typically 2D for infovis
• keep/explain as much variance as possible
• show underlying dataset structure
• multidimensional scaling (MDS)
• minimize differences between interpoint distances
  in high and low dimensions

21
Dimensionality Reduction Example

• 4096D pixels in image
• 2D 2 new axes represent wrist rotation and
  finger extention

A Global Geometric Framework for Nonlinear
Dimensionality Reduction. Tenenbaum, de Silva,
and Langford. Science 290 (5500), pp 2319--2323,
Dec 22 2000
22
Glimmer Multi-Level MDS on the GPU

• speed through GPGPU parallelism
• exploit commodity graphics cards
• multi-level approach to avoid slowdown or
  incorrect termination in local minima

Glimmer
GPU-SF
Hybrid
joint work with Stephen Ingram, Marc Olano (UMBC)
23
Sparse Example

• automatically finding correct spatial structure
  to match human-generated clusterings (colored)

Glimmer
Hybrid
Landmark
24
Glimmer Speed
25
Glimmer Speed
26
Glimmer Speed Detail
27
Glimmer Speed
28
Glimmer vs. GPUSF Detail
29
Glimmer Video
30
Current Research

• accordion drawing
• started with evolutionary trees, gene sequences
• monitoring large collections of machines
• log visualization
• analyzing large session logs
• dimensionality reduction
• finding high-dimensional clusters quickly using
  GPU
• network visualization
• general multi-level graph drawing
• specialized protein-protein interaction networks
• infovis evaluation
• understanding when and how these techniques work
  as planned

31
Multi-Level Graph Drawing

• TopoLayout
• multi-level
• decompose and lay out by topological features

joint work with Dan Archambault, David Auber
(Bordeaux)
32
Grouse Interactive Hierarchy Exploration
joint work with Dan Archambault, David Auber
(Bordeaux)
33
Protein Interaction Diagrams

• Hand drawn diagrams
• cellular location encoded spatially
• activated proteins grouped by function

34
Cerebral Protein Interaction Networks

• Automatic drawing using hard and soft constraints
  inspired by hand drawn diagrams

joint work with Aaron Barsky, Jennifer Gardy (UBC
Microbiology)
35
Cerebral Results MAPK network

• N760, E1263. Time 77 seconds

36
Previous Work IpSep-Cola
37
Previous Work GEM
38
Current Research

• accordion drawing
• started with evolutionary trees, gene sequences
• monitoring large collections of machines
• log visualization
• analyzing large session logs
• dimensionality reduction
• finding high-dimensional clusters quickly using
  GPU
• network visualization
• general multi-level graph drawing
• specialized protein-protein interaction networks
• infovis evaluation
• understanding when and how these techniques work
  as planned

39
Evaluation with HCI and Psych Collab

• Lau, Rensink, and Munzner. Perceptual Invariance
  of Nonlinear FocusContext Transformations. Proc
  APGV 04, p 65-72.
• Lam, Rensink, and Munzner. Effects of 2D
  Geometric Transformations on Visual Memory. Proc
  APGV 06, p 119-126.
• evaluating effect of nonlinear distortion on
  visual search/memory
• Nekrasovski, Bodnar, Guimbretiere, McGrenere, and
  Munzner. An Evaluation of PanZoom and Rubber
  Sheet Navigation with and without an Overview.
  Proc CHI 2006, p 11-20.
• evaluating some aspects of accordion drawing

40
Evaluation with HCI and Psych Collab

• Lam, Munzner, and Kincaid. Overview Use in
  Multiple Visual Information Resolution
  Interfaces. To appear Proc. InfoVis 07, published
  as IEEE TVCG Nov/Dec 2007.
• funded by Agilent, followup after Lam internship
• evaluating techniques for exploring large
  collections of time-series data

41
More Information

• papers
• http//www.cs.ubc.ca/tmm/papers.html
• talks
• http//www.cs.ubc.ca/tmm/talks.html