UCLA Human Perception Laboratory

1
UCLA Human Perception Laboratory
Collaborators
Jerry Balakrishnan, Purdue University
Tim Clausner, HRL Laboratories
Thomas Shipley, Temple University
Thomas Wickens, UCLA
Students
Timothy Burke, UCLA
Julia Cohen, UCLA
Patrick Garrigan, UCLA
Sharon Guttman, UCLA
Evan Palmer, UCLA
2
User Interface Aspects of Decision Aid
Systems 1) Optimizing Interfaces 2)
Designing for Human Decisionmaking 3)
Optimizing the Operator Training
3
Recent Project-Relevant Publications
Book
Shipley, T.F. Kellman, P.J. (Eds.) (2001).
From fragments to objects Segmentation and
grouping in vision. Elsevier Science
Press.
Articles
Kellman, P.J. (in press). Vision - occlusion,
illusory contours and 'filling in." In
Encyclopedia of Cognitive Science, Oxford,
UK Nature Publishing Group.
Kellman, P.J. (2001). Separating processes in
object perception. Journal of Experimental Child
Psychology, 78, 84-97.
Yin, C., Kellman, P.J. Shipley, T.F. (2000).
Surface integration influences depth
discrimination. Vision Research,
40(15), 1969-1978.
Chapters
Kellman, P.J. (in press). Segmentation and
grouping in object perception A 4- dimensional
approach. To appear in M. Behrmann and
R. Kimchi (Eds.). Object perception The 31st
Carnegie-Mellon Symposium on
Cognition. Hillsdale, NJ Erlbaum.
Kellman, P.J. (2002). Perceptual learning. In
R. Gallistel (Ed.), Stevens' handbook of
experimental psychology, Third edition,
Vol. 3 (Learning, motivation and emotion), John
Wiley Sons.
Kellman, P.J., Guttman, S. Wickens, T.
(2001). Geometric and neural models of contour
and surface interpolation in visual object
perception. In Shipley, T.F. Kellman, P.J.
(Eds.) From fragments to objects Segmentation
and grouping in vision. Elsevier
Science Press.
Patent Applications Submitted
System and Method for Adaptive Learning. US
10,020,718. Inventor P. Kellman. Filed by
Kellman A.C.T. Services, Inc.
System and Method for Representation of
Aircraft Altitude using Natural Perceptual
Dimensions. Inventors P. Kellman, T.
Clausner, E. Palmer. Filed by Raytheon Company.
4
OptimizingInterfaces

5
THE QUANDARY OF MULTIMEDIA AND MULTICHANNEL
COMMUNICATIONS
Ten or more simultaneous voice channels
Computer Monitors
Several Video Displays
Pilot Copilot Chase Pls Radar Observers Engineers
Flight Controller NAVY, NASA
6
S
S
Decision Aid Suggestions, Estimates
Resource Information
S
Communications / Collaborative Inputs
S
System status information
S
Primary situation displays
7
Principles of Perception and Attention

• Acuity, resolution limits
• Contrast
• Color
• Motion
• Grouping and Segmentation
• Pop-out
• Highlighting Important Relations

8
Find an L
9
(No Transcript)
10
Find an X
11
(No Transcript)
12
Raise your hand when any characterpops out.
13
(No Transcript)
14
There are general principles of perception
15
There are general principles of perception
Are there general principles for effective
interfaces?
16
There are general principles of perception
Are there general principles for effective
interfaces?
Strategy Look at several real-world tasks,
their current interfaces, and
potential improvements
17
Recurring Themes
Crucial Importance of Task Analysis
18
Recurring Themes
Crucial Importance of Task Analysis
Tradeoff Between Clutter and Navigation
Demands
19
Example Water System Management
20
(No Transcript)
21
Service Details
22
Recurring Themes
Crucial Importance of Task Analysis
Tradeoff Between Clutter and Navigation
Demands
Balance between Familiar and Improved
23
Example Air Traffic Control
24
Example Air Traffic Control
Coding using natural perceptual dimensions
25
(No Transcript)
26
No Cues
Size
Gray
Size Gray
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
Recurring Themes
Crucial Importance of Task Analysis
Tradeoff Between Clutter and Navigation
Demands
Balance between Familiar and Improved
Separating Channels
33
AN IMPROVED MISSION-CRITICAL INTERFACE Virtual
Audio Each speech channel appears to come
from a different location in 3-D.

Recent Progress Automatic keyword and voice
recognition Protection from cognitive illusions
34
Recurring Themes
Crucial Importance of Task Analysis
Tradeoff Between Clutter and Navigation
Demands
Balance between Familiar and Improved
Separating Channels
The Bells and Whistles Problem
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
107.3 PSI 1593.0 HG
39
Recurring Themes
Crucial Importance of Task Analysis
Tradeoff Between Clutter and Navigation
Demands
Balance between Familiar and Improved
Separating Channels
The Bells and Whistles Problem
Conveying the Big Picture
40
(No Transcript)
41
? Efficient Interfaces
Integrating Human Decisionmaking with
? Intelligent Systems
42
Cognitive Illusions

• Result from use of invalid heuristics rather than
  normative
• models

• Can affect even experts

• Exacerbated by time pressure, stress

• Reflect a tradeoff heuristics give us quick,
  good answers
• most of the time

43
THE THEORY OF MENTAL MODELS
Prof. Philip Johnson-Laird -- Princeton

• Could serve as basis of software to detect
• likely fallacies of human reasoning

44
Long-range Task Create software that
anticipates human decisionmaking illusions.
More Tractable Tasks Improve human decisions
with decision aids and efficient interfaces to
utilize them fully. Use assessment and
training to identify and remedy characteristic
errors.
45

• Listen For Warning Sounds
• Traffic Alert
• Manually Click on Satellite Icon to perform
  Orbital Recovery Reinitialize
• Establish Audio Communication to Acknowledge
  the Alarm
• ("Initialize, One, High)
• Or Establish Audio Communication to Ignore
• Malfunction Alert
• Manually Correct the Satellite Path Rescue
  Fix
• Establish Audio Communication to Recovery
  Rescue
• (Rescue, Two, Low)
• Or Establish Audio Communication to Ignore
• Satellite Maneuvering to Avoid Restricted Areas
• User-Dependent Probabilistic Events
• Event probabilities depends on operators
  interaction with display and risk state
• Establish Audio Communication to Enter or
  Avoid
• (Avoid, Two, 80)
• Manually Correct the Satellite

46
Optimizing the OperatorTraining for
Recognition,Decisionmaking,and Action

47
Optimizing the OperatorTraining

Information Extraction
48
Optimizing the OperatorTraining

Information Extraction
Procedure Execution
49
Optimizing the OperatorTraining

Information Extraction
Procedure Execution
Decisionmaking
50
Learning Techniques

• Information Extraction
• Perceptual Learning Methods

51
Learning Techniques

• Information Extraction
• Perceptual Learning Methods
• Decisions and Procedures
• Challenge - Response Paradigms

52
Why not just have a nice tutorial?
53
Why not just have a nice tutorial?
Limitations of declarative knowledge
54
Why not just have a nice tutorial?
Limitations of declarative knowledge
Intuitive knowledge is implicit
55
Why not just have a nice tutorial?
Limitations of declarative knowledge
Intuitive knowledge is implicit -- not
necessarily verbalizable
56
Why not just have a nice tutorial?
Limitations of declarative knowledge
Intuitive knowledge is implicit -- not
necessarily verbalizable -- involves perceptual
classification, pattern recognition
57
Why not just have a nice tutorial?
Limitations of declarative knowledge
Intuitive knowledge is implicit -- not
necessarily verbalizable -- involves perceptual
classification, pattern recognition -- grows
by experience
58
Perceptual Learning
? Changes in the information extraction as a
result of practice or experience
59
NOVICE
EXPERT

SEARCH TYPE
Serial Processing
Parallel Processing
Selective Attention to Relevant Information
Filtering Out of Irrelevant Information
FILTERING
Attention to Irrelevant and Relevant
Information
UNITS
Low-Level Features
Chunks / Higher-order Relationships
ATTENTIONAL LOAD
High
Low
Slow
Fast
SPEED
CONTROLLED PROCESSING
AUTOMATIC PROCESSING
60
How Do We Learn Classifications and Structures
Naturally?
61
How Do We Learn Classifications and Structures
Naturally?

• Not through tutorials

62
How Do We Learn Classifications and Structures
Naturally?

• Not through tutorials

• Encounter examples and discover similarities
• (unsupervised learning)

63
How Do We Learn Classifications and Structures
Naturally?

• Not through tutorials

• Encounter examples and discover similarities
• (unsupervised learning)

• Encounter examples and receive feedback
• (supervised learning)

64
How Do We Learn Classifications and Structures
Naturally?

• Not through tutorials

• Encounter examples and discover similarities
• (unsupervised learning)

• Encounter examples and receive feedback
• (supervised learning)

• Learning grows by classification experience

65
How Do We Learn Classifications and Structures
Naturally?
66
How Do We Learn Classifications and Structures
Naturally?
67
How Do We Learn Classifications and Structures
Naturally?
cat
68
How Do We Learn Classifications and Structures
Naturally?
cat
dog
69
How Do We Learn Classifications and Structures
Naturally?
cat
dog
70
How Do We Learn Classifications and Structures
Naturally?
cat
dog
71
How Do We Learn Classifications and Structures
Naturally?
cat
dog
72
How Do We Learn Classifications and Structures
Naturally?
cat
dog
73
How Do We Learn Classifications and Structures
Naturally?
cat
dog
74
How Do We Learn Classifications and Structures
Naturally?
cat
dog
75
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification

Several features of PLMs are patent pending,
Insight Learning Technology, Inc. For use in
your application, contact .
76
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials

77
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation

78
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation
• Modeled as a Filtering Process

79
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation
• Modeled as a Filtering Process
• Discovery of relevant features and relationships

80
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation
• Modeled as a Filtering Process
• Discovery of relevant features and relationships
• Suppression of irrelevant detail

81
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation
• Modeled as a Filtering Process
• Leads to Automaticity

82
Accelerated ExpertisePerceptual Learning
Modules -- PLMs

• Speeded Classification
• Many Short Trials
• Invariant Structure within Changing Irrelevant
  Variation
• Modeled as a Filtering Process
• Leads to Automaticity
• Transforms some Decisionmaking into Pattern
  Recognition

83
(No Transcript)
84
(No Transcript)
85
(No Transcript)
86
(No Transcript)
87
(No Transcript)
88
(No Transcript)
89
PLMs Multiple Levels

• Learning interface symbols and layout

90
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?

91
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations

92
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations
• What is happening now?

93
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations
• What is happening now?
• Procedures for obtaining added information

94
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations
• What is happening now?
• Procedures for obtaining added information
• How do I get a probability estimate for event X?

95
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations
• What is happening now?
• Procedures for obtaining added information
• How do I get a probability estimate for event X?
• Making decisions based on arrays of information

96
PLMs Multiple Levels

• Learning interface symbols and layout
• Where do I look for X?
• Recognition and classification of situations
• What is happening now?
• Procedures for obtaining added information
• How do I get a probability estimate for event X?
• Making decisions based on arrays of information
• What do we do?

97
Procedures Training

• Shares Characteristics of Perceptual
  Learning Modules

98
Procedures Training

• Shares Characteristics of Perceptual
  Learning Modules
• Challenge - Response Trials

99
Procedures Training

• Shares Characteristics of Perceptual
  Learning Modules
• Challenge - Response Trials
• Objective Assessment

100
Procedures Training

• Shares Characteristics of Perceptual
  Learning Modules
• Challenge - Response Trials
• Objective Assessment
• Learning to Criterion

101
Procedures Training

• Shares Characteristics of Perceptual
  Learning Modules
• Challenge - Response Trials
• Objective Assessment
• Learning to Criterion

102
(No Transcript)
103
Recent Progress Adaptive Learning Using
Novel Sequencing Algorithms
104
Sequencing Algorithms
Very general they can be applied to
Perceptual Learning (classification,
recognition) Procedure Learning Any Set
of Memory Items
105
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
106
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
Rapid reappearance of missed items
107
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
Rapid reappearance of missed items
Assess nature of cognitive processing
(deliberative / slow vs. automatic /
fast)
108
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
Rapid reappearance of missed items
Assess nature of cognitive processing
(deliberative / slow vs. automatic /
fast)
Stretch the retention interval as learning
improves
109
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
Rapid reappearance of missed items
Assess nature of cognitive processing
(deliberative / slow vs. automatic /
fast)
Stretch the retention interval as learning
improves
Retire learned items
110
Components of Optimal Sequencing
Exercise encoding and retrieval in LTM, not
STM
Rapid reappearance of missed items
Assess nature of cognitive processing
(deliberative / slow vs. automatic /
fast)
Stretch the retention interval as learning
improves
Retire learned items
Speed, accuracy and durability criteria
111
(No Transcript)
112
Intersection Angle (deg)
113
(No Transcript)
114
(No Transcript)
115
(No Transcript)
116
Variation among Instances of a Concept
Orientation
Altitude
Context
117
Negative instances may also be sequenced.
118
Priority Scores
119
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
120
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
All items assigned initial priority scores
121
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
All items assigned initial priority scores
Uniform assignment
122
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
All items assigned initial priority scores
Uniform assignment
leads to random selection
123
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
All items assigned initial priority scores
Uniform assignment
Scaffolding
124
Priority Scores
Every item in the learning set has a priority
that is updated after each learning trial
All items assigned initial priority scores
Uniform assignment
Scaffolding
leads to presentation of certain
key items or concepts first
125
Sequencing Algorithm
Uses subjects performance to arrange the
sequence of items or concept types
Cuts learning time roughly in half in all domains
studied so far
Sequencing algorithm is patent pending, Insight
Learning Technology, Inc. For use in your
application, contact .
126
Sequencing Techniques
-- Extremely general items, complex
classifications, procedures
127
Sequencing Techniques
Research Questions
Best parameters for speed, durability of
learning
Do best parameter settings for item learning
differ from perceptual and procedural
learning?
128
Summary of Sequencing

• Adaptive method optimizes several principles of
  learning
• Drastically reduces learning time
• Objective measures of speed and accuracy
• Effort is directed where it is needed most
• Learned items are retired
• Allows learning to objective criteria
• More comprehensive learning (all concepts or
  items)
• Effective for initial or recurrent training

129
Recent Project-Relevant Publications
Book
Shipley, T.F. Kellman, P.J. (Eds.) (2001).
From fragments to objects Segmentation and
grouping in vision. Elsevier Science
Press.
Articles
Kellman, P.J. (in press). Vision - occlusion,
illusory contours and 'filling in." In
Encyclopedia of Cognitive Science, Oxford,
UK Nature Publishing Group.
Kellman, P.J. (2001). Separating processes in
object perception. Journal of Experimental Child
Psychology, 78, 84-97.
Yin, C., Kellman, P.J. Shipley, T.F. (2000).
Surface integration influences depth
discrimination. Vision Research,
40(15), 1969-1978.
Chapters
Kellman, P.J. (in press). Segmentation and
grouping in object perception A 4- dimensional
approach. To appear in M. Behrmann and
R. Kimchi (Eds.). Object perception The 31st
Carnegie-Mellon Symposium on
Cognition. Hillsdale, NJ Erlbaum.
Kellman, P.J. (2002). Perceptual learning. In
R. Gallistel (Ed.), Stevens' handbook of
experimental psychology, Third edition,
Vol. 3 (Learning, motivation and emotion), John
Wiley Sons.
Kellman, P.J., Guttman, S. Wickens, T.
(2001). Geometric and neural models of contour
and surface interpolation in visual object
perception. In Shipley, T.F. Kellman, P.J.
(Eds.) From fragments to objects Segmentation
and grouping in vision. Elsevier
Science Press.
Patent Applications Submitted
System and Method for Adaptive Learning. US
10,020,718. Inventor P. Kellman. Filed by
Kellman A.C.T. Services, Inc.
System and Method for Representation of
Aircraft Altitude using Natural Perceptual
Dimensions. Inventors P. Kellman, T.
Clausner, E. Palmer. Filed by Raytheon Company.
130
(No Transcript)