Transcript No Slide Title

May 11, 2006
IBM Almaden Institute
Cognitive Computing
Cortical Dynamics of
Working Memory
Joaquín M. Fuster
Semel Institute for Neuroscience
and Human Behavior
University of California
Los Angeles
www. joaquinfuster. com

Working memory is the active (“online”)
retention of information for a prospective
action to solve a problem or to reach a goal (vs
STM).

That information is unique for present
context and for the immediate future, but
inseparable from past context: It is long-term
memory updated for prospective use.

Working memory and long-term memory
share most of the same cortical structure.
A corollary:
A sensory stimulus to be retained in working
memory activates a cortical network that
encodes not only that stimulus, but also its
past associative context in long-term memory.
1. Structure of cortical memory networks
2. Cortical dynamics of working memory (WM)
The Perception-Action Cycle
(a) Cross-modal integration by WM
(b) Neurovascular coupling in WM
3. Computational constraints
4. WM dynamics in the human cortex
A NETWORK PARADIGM
1. Memory and knowledge are represented
in widely distributed, interactive, and
overlapping neuronal networks of the cerebral
cortex (cognits).
2. Cognitive functions--perception, attention,
memory, including working memory, language,
and intelligence--are based on neural
transactions within and between those cortical
memory networks.
Memories (cognits) consist of widely distributed networks
made of neurons synaptically modulated by experience
Executive memory
in frontal cortex
Perceptual memory
in posterior cortex
Networks intersect and overlap with one another profusely: A
neuron can be part of many networks, thus many memories
Hebb 1949
Carew et al 1984
Scoville & Milner 1957
Corkin 1984
Squire 1987
Cohen & Eichenbaum 1993
Limbic structures essential
for memory in neocortex
Order of maturation
of cortical areas
Gibson 1991
Sowell et al 1999
Bartzokis 2003
Gillery 2005
Bonin, G. von, Essay
on
the Cerebral Cortex,
C.C. Thomas, 1950
Evolution of frontal
association cortex
(prefrontal)
 Maturation progresses from primary areas to
progressively higher association areas.
 “Upstream” connectivity also progresses from
primary areas to progressively higher association
areas.
Jones & Powell 1970
Pandya & Yeterian 1985
Cavada & Goldman-Rakic 1989
Felleman & Van Essen 1991
 Thus, three structural gradients from primary to
association cortex :
a) Evolution
b) Maturation
c) Connectivity
 Along those gradients, memory networks
(cognits) become layered hierarchically, from
concrete sensory and motor memory at the bottom
to abstract knowledge at the top.
 Every new memory follows those gradients as it
becomes associated with concurrent stimuli and, at
higher levels, with pre-existent memories.
Lorente de Nó 1938
Jones & Powell 1970
Pandya & Yeterian 1985
Cavada & Goldman-Rakic 1989
Felleman & Van Essen 1991
Memory formation in networks of the cerebral cortex, from
sensory up to association cortex, through 11 cells or groups
of cells (3 layers) linked by typical patterns of connectivity
(feed-forward, feed-back, convergence, divergence, lateral)
Executive memory
Perceptual memory
Executive memory
Perceptual memory
Executive memory
Perceptual memory
Executive memory
Perceptual memory
Executive memory
Perceptual memory
At all hierarchical levels, executive and
perceptual networks are associatively connected,
thus executive memories contain perceptual elements
or assemblies, and vice versa.
This executive-perceptual connectivity becomes
critical in the integrative dynamics of working
memory.
Essential to the dynamics of working memory, to
ensure the orderly pursuit of a goal, is the mediation of
cross-temporal contingencies between percepts and
actions:
IF NOW THIS, THEN LATER THAT;
IF EARLIER THAT, THEN NOW THIS.
J. von Uexküll, Theoretical Biology, Harcourt, 1926
Working Memory
Weizsäcker 1950
Neisser 1976
Arbib 1981
Cortical Perception-Action Cycle
Inputs to frontal executive networks at
the top of the Perception-Action Cycle
Outputs from frontal executive
networks at the top of the PerceptionAction Cycle
Working Memory
Whereas the cycle
operates in series and in
parallel through the
environment, integrative
working memory at the top
operates by reentrant
cortical integration (RCI).
Edelman, G.M. The Remembered Present. A Biological Theory of
Consciousness. New-York: Basic Books, 1989.
Fuster, J.M. and Alexander, G.E., Science 173:652-654. 1971
PREFRONTAL
Niki 1974
Fuster et al. 1982
Quintana et al. 1988
Barone & Joseph 1989
Funahashi et al. 1989
Quintana & Fuster 1989
Requin et al. 1990
Fuster et al. 2000
TEMPORAL
Fuster & Jervey 1982
Fuster et al. 1985
Miyashita 1988
Miyashita & Chang 1988
Miller et al. 1993
Tomita et al. 1999
PARIETAL
Gnadt & Andersen 1988
Quintana et al. 1989
Andersen & al. 1990
Barash & Andersen 1991
Quintana & Fuster 1999
Zhou & Fuster 1996
Chafee & Goldman-Rakic 2000
Ardestani & al. 2005
Cross-Modal Temporal Integration in
Working Memory
Fuster et al., Nature 405:347-351, 200
Fuster et al., Nature 405:347-351, 2000
Zhou,Y. and Fuster, J.M., PNAS 93:10533-10537,199
Neurovascular Coupling
in Working Memory
Near-Infrared Spectroscopy (NIRS)
Surface Field Potentials
Local Field Potentials
Unit Discharge
PARIETAL
FRONTAL
5
PS
9
8
AS
7a
IS
10 mm
LS
STS
The specialized WM “modules” of
prefrontal cortex are nodes of heavy
association in active executive-memory
networks.
Those nodes encode specific sensorymotor associations within widely distributed
cortical networks, networks that encode the
broad behavioral context in which those
specific associations were formed and
remain embedded.
Cognit Modeling Constraints
Architecture (network)
Relational code
Hierarchical & heterarchical
Plasticity
Dynamics
Content-addressable
Stochastic
Reentry
Hierarchical & heterarchical
Parallel
Serial
Categorical (degeneracy)
(a) in perception
(b) in action
Cortical Working Memory
2006
Art work:
Amanda Hammond
Kim Hager
UCLA Brain Mapping
Consultants:
Joaquín Fuster
Allen Ardestani
UCLA Semel Institute
Cognitive Neuroscience
Primate Units
Fuster, J. Neurophysiol. 36: 61 (1973)
Niki & Watanabe, Brain Res. 171: 213
(1979).
Fuster & Jervey, J. Neurosci. 2: 361
(1982).
Kojima & Goldman-Rakic, Brain Res.
248: 43 (1982).
Fuster et al., Brain Res. 330: 299
(1985).
Funahashi et al., J. Neurophysiol. 61:
331 (1989).
Quintana et al., Brain Res. 503:100
(1989).
Miller et al., J. Neurosci. 13: 1460 (1993).
Quintana & Fuster, Cerebral Cortex 9:
213 (1999).
Chafee & Goldman-Rakic, J. Neurophysiol. 83: 1550 (2000).
Fuster et al., Nature 405: 347 (2000).
Schultz et al., Cerebral Cortex 10: 272
(2000).
Human Imaging
Courtney et al., Nature 386: 608 (1997).
Petit et al., J. Neurosci. 18: 9429 (1998).
D’Esposito et al., Exp. Brain Res. 133: 3
(2000).
Pollmann & Von Cramon, Exp. Brain
Res. 133: 12 (2000).
*Duncan & Owen, TINS 23: 475 (2000).
*Cabeza & Nyberg, J. Neurosci. 12: 1
(2000).
Mecklinger et al., Hum. Brain Mapp.
11: 146 (2000).
*Wager & Smith, Cog. Affect. Beh.
Neurosci. 3: 255 (2003).
Crottaz-Herbette et al., NeuroImage
21: 340 (2004).
Buchsbaum et al., Neuron 48: 687
(2005).
Goldstein et al., Neuropsychology 19:
509 (2005).
*Rajah & D’Esposito, Brain 128: 1964
(2005).
Superior
parietal
SMA
Superior
frontal
Anterior
cingulate
Middle
frontal
Inferior
frontal
Motor
Fusiform
Superior
temporal
Inferior
temporal
Orbital
Time
Inferior
parietal
Visual WM
Visual
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
Superior
parietal
SMA
Superior
frontal
Anterior
cingulate
Middle
frontal
Inferior
frontal
Motor
Fusiform
Superior
temporal
Inferior
temporal
Orbital
Time
Inferior
parietal
SPATIAL WM
Visual
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
Superior
parietal
SMA
Superior
frontal
Anterior
cingulate
Middle
frontal
Inferior
frontal
Motor
Fusiform
Superior
temporal
Inferior
temporal
Orbital
Rabbit Apple
Virtue Pen
Bread Music
Pen
Time
Inferior
parietal
VERBAL WM
Visual
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
Conclusions
1. The memory code is a relational code: memories
are networks made of associations between more or less
dispersed neuronal assemblies of the neocortex.
2. Perceptual memory networks are organized
hierarchically in posterior cortex; executive memory networks
in frontal cortex.
3. The prefrontal cortex, at the top of the perceptionaction cycle, mediates cross-temporal contingencies with its
integrative executive function of working memory.
4. Working memory is maintained by recurrent activity
between prefrontal cortex and associative areas of posterior
cortex.
Yong-Di Zhou
Mark Bodner
James Kroger
Allen Ardestani
Michael Guiou
Arthur Toga