ZhongLin Lu

1
Perceptual Learning
Zhong-Lin Lu Dana David Dornsife Cognitive
Neuroscience Imaging Ceter Departments of
Psychology Biomedical Engineering
Neuroscience Graduate Program University of
Southern California Supported by NSF, NIMH
and NEI
2
George Stratton (1896) UC Berkeley

• Inverted lenses
• The world looks upside down

A modern version of the lenses
3
At first

• Things were seen in one way and thought of in a
  far different way

4
Finally, the 8th day, lenses removed

• The 3rd day
• Walking required much less care
• The 5th day
• Things seemed almost normal
• The 7th day
• Enjoyed for the first time the scene in the
  evening walk

• For several hours things were upside down

5
What is perceptual learning?

• Improved behavioral performance as a result of
  PERCEPTUAL experiences
• More examples
• Wine tasting
• Reading x-rays
• Prism goggles

A General Conclusion Perception and perceptual
learning cant be studied separately --- the
adult perceptual system has a very high degree of
plasticity!
6
Psychophysics/physiological studies with simple,
well-controlled, lab stimuli

• Learning specificity
• Law of learning
• Mechanisms of learning
• Level and Mode of learning
• Optimal training methods

7
Trademark findings
1. We can learn 2. Learning is stimulus specific
performance
time
Learning specificity
8
No transfer when rotating Vernier 90o
correct
Session
(Poggio, Fahle, Edelman, 1992)
Learning specificity
9
Perceptual Learning in behavioral assay in many
perceptual tasks in humans

• Detection or discrimination of visual gratings or
  patterns (DeValois, 1977 Mayer, 1983
  Fiorentini Berardi, 1980, 1981 Fine
  Jacobs,2000Dorais Sagi, 1997)
• Stimulus orientation judgments (Vogels Orban,
  1985 Shiu Pashler, 1992 Schoups, Vogels
  Orban, 1995 Dosher Lu, 1998 1999 Matthews,
  Liu Qian, 2001 Rivest, Boutet Intriligator,
  1997)
• Motion direction discrimination (Ball Sekuler,
  1982, 1987 Ball, Sekuler, Machamer, 1983 Liu
  Vaina, 1998 Matthews, et al, 1999 Wehrhahn
  Rapf, 2001 Zanker, 1999) KDE (Vigyasagar
  Stuart, 1994)
• Texture discrimination and visual search (Karni
  Sagi, 1991, 1993 Ahissar Hochstein, 1996,
  1997 Schoups Orban, 1996 Ahissar, et al,
  1998Rubenstein Sagi, 1993 Sireteanu
  Rettenbach, 1995, 2000)
• Time to perceive random dot stereograms
  (Ramachandran Braddick, 1973) or Stereoacuity
  (Fendick Westheimer, 1983)
• Hyperacuity and vernier acuity (Beard, Levi,
  Reich, 1995 Bennett Westheimer, 1991 Fahle
  Edelman, 1993 Fahle, Edelman Poggio, 1995
  Kumer Glaser, 1993 McKee Westheimer, 1978
  Saarinen Levi, 1995 De Luca Fahle, 1999
  Herzog Fahle, 1997, 1999 Westheimer, 2001)
• Object and face recognition (Gold, Bennet
  Sekular, 1999 Furmanski Engel, 2000).

Learning specificity
10
Behavioral specificities in perceptual learning

• Retinal location (Ramachandran Braddick, 1973
  Karni Sagi, 1991 1993 Schoups, Vogels
  Orban, 1995 Dosher Lu, 1999 Nazir ORegan,
  1990 OToole Kersten, 1992 Kapadia, Gilbert
  Westheimer, 1994 Shiu Pashler, 1992 Ball
  Sekuler, 1982, 1987)
• Orientation, spatial frequency, object set
  (Fiorentini Berardi, 1980, 1981 Mayer, 1983
  Ramachandran Braddick, 1973 Karni Sagi,
  1991 Saarinen Levi, 1995 Fahle Edelman,
  1993 Schoups, Vogels Orban, 1995 Kapadia,
  Gilbert Westheimer, 1994 Dorais Sagi, 1997
  Furmanski Engel, 2000)
• Motion direction (Ball Sekuler, 1982, 1987 De
  Luca Fahle, 2001)
• Task (Fahle, Edelman Poggio, 1995 Zanker,
  1999 Fahle, 1997 Ball Sekuler, 1982, 1987
  Matthews, et al, 1999)
• Configuration and/or stimulus attribute (Kapadia,
  Gilbert Westheimer, 1994 Dorais Sagi, 1997
  Ahissar Hochstein, 1993 Karni Sagi, 1991
  Sagi Palot, 1992 Shiu Pashler, 1992
  Rubenstein Sagi, 1993)
• Perceptual learning ? cognitive learning,
  strategy selection, or motor learning

Learning specificity
11
Dosher Lu, 2007
Law of Learning
12
Dosher Lu, 2007
Law of Learning
13
The Equivalent Input Noise Psychophysics
Mechanisms of Learning
14
PTM Mechanisms of State Change
Multiplicative Noise Reduction
Mechanisms of Learning
Lu Dosher, 1998
15
Mechanism analogies in signal processing,
physiology, and PTM

• Signal Processing Physiology Psychophysics
• Amplification Gain increase Stimulus
  enhancement
• 
  
• Filtering Retuning External noise exclusion
• 
  
• Gain Control Nonlinearity nonlinear
  ity/ mult noise reduction
• 
  
• Gain change without tuning changes Luck
  Mangun 1995 McAdams Maunsell, 1999
• Neural retuning (altering sensitivity profiles)
  Moran Desimone, 1994 Treue, Maunsell,
  Trujillo, 1999 Patzwahl et al, 1998
• Non-linearity in response, and modified
  non-linearitiesHeeger, 1995 Lee et al, 1998,
  model

Mechanisms of Learning
16
Perceptual Learning at 8 Noise Levels and Two
Criteria
Dosher Lu PNAS 1998
Mechanisms of Learning
17
Representational Enhancement or Connection
Reweighting

• Perceptual learning may be due to recruitment of
  new units or sharpening the existing ones.
• Dominant hypothesis is modification of early
  representation

Level and Mode of Learning
18
Selective Re-Weighting
Representation and Connection Hebbian Learning
Level and Mode of Learning
19
Learning to weight representations with better
signal to noise value
Level and Mode of Learning
20
Level and Mode of Learning
21
Perceptual learning in clear displays
optimizes perceptual expertise
Dosher Lu, 2005
Optimal Training Methods
22
Lu, Chu Dosher, 2006
Optimal Training Methods
23

• Importance of REM sleep in some perceptual
  learning tasks (Karni Sagi, 1989).
• Trade-off between perceptual fatigue and
  perceptual learning (Censor Sagi, 2008) and the
  benefit of taking a nap between training sessions
  (Mednick, Nakayama, Cantero, Atienza, Levin,
  Pathak, Stickgold, 2002).

Optimal Training Methods
24
Liu Weinshall, 2000
Optimal Training Methods
25
Adaptive Methods

• Adaptive methods are well-known in psychophysics.
• The idea is to use dynamic stimulus placement
  strategies based on subjects responses to
  optimize the efficiency of data collection.
• Development of adaptive methods has mostly
  focused on psychometric functions.

Our goal is to develop and test adaptive methods
for characterizing other psychological functions.
Optimal Training Methods
26
Adaptive Methods The general Idea

• Four components
• Characterization of the psychological function
  with a functional form.
• Trial-by-trial update of the estimate of the
  parameters of the functional form.
• Selection of the optimal stimulus condition.
• Re-iteration and a stop rule.

Optimal Training Methods
27
Bayesian Update

• Define a parameter space T? with a prior
  distribution p0(?)? where ? is a
  multi-dimensional vector representing the
  parameters of the psychological function. The
  entire parameter space represents all the
  possible psychological functions???
• Use Bayes rule to evaluate the posterior
  probability, pt (?) following the observers
  response to stimulus xt in trial t.
• Calculate pt1(???for all possible stimulus
  conditions and responses, and the expected
  entropy of pt1(?? before running trial t1.
• On trial t1, present to the observer the
  stimulus condition providing the minimum expected
  entropy for pt1(?).

(Cobo-Lewis, 1996Kontsevich Tyler, 1999
Lesmes, et al, 2006)
Optimal Training Methods
28
qTvC
Optimal Training Methods
29
qCSF
Optimal Training Methods
30
Applications to older adults at risk for AD (ADRC
_at_ USC Clark Lu Pfizer _at_ Kentucky Carbary,
Mullineaux, Schmitt, Lu)
Optimal Training Methods
31
Other qMethods
qSAT qFTvC qColorEllipse qPerimetry qfMRI
Optimal Training Methods
32
Optimal Training Methods
33
Question What kinds of mission-related tasks can
we apply the theoretical and empirical knowledge
of perceptual learning to?
Theory Observer/Operator learning
models Research Questions Optimal training
schedule Optimal training environment Technology
Adaptive learning protocols Quick and
non-invasive assessment
So what?
34

• Research Supported by
• National Institutes of Mental Health
• National Science Foundation
• Air Force Office of Scientific Research