cognitive neuroimaging, predictive coding, and unifying ... · at multiple levels of analysis...
TRANSCRIPT
1
Cognitive Neuroimaging,
Predictive Coding,
and Unifying Theories of the BrainLars Muckli
Centre for Cognitive Neuroimaging (CCNi)
Department of Psychology, University of Glasgow, UK
1
Edinburgh – 16 June 2010
2
Cognitive Neuroimaging at the CCNi - Glasgow
2
3T – MRI Siemens Tim
32 channel
TMS - EEG128 channel
MR compatible
MEG –
Biomag 4D248-channel
magnetometer
3
CCNi Research
Principle investigators:
Pascal Belin
Roberto Caldara
Simon C. Garrod
Marie-Hélène Grosbras
Joachim Gross
Klaus Kessler
people
…to advance the understanding of
the complex relationship between
the brain, cognition and behaviour
at multiple levels of analysis
mission
Principle investigators:
Pascal Belin
Roberto Caldara
Simon C. Garrod
Marie-Hélène Grosbras
Joachim Gross
Klaus Kessler Klaus Kessler
Hartmut Leuthold
Lars Muckli
Guillaume A. Rousselet
Philippe G. Schyns
Gregor Thut
Klaus Kessler
Hartmut Leuthold
Lars Muckli
Guillaume A. Rousselet
Philippe G. Schyns
Gregor Thut
4
Voice Neurocognition LaboratoryPascal Belin
5
Cultural Neuroscience
We aim at mapping human diversity and understanding how culture
and race modulate visual and social cognition, by using a
multidisciplinary approach
BRAIN DAMAGED PATIENTS
Roberto Caldara
EYE MOVEMENTSEEG
fMRICOMPUTATIONAL MODELLING
6
Single-trial parametric GLM of EEG data
Guillaume A. Rousselet & Cyril R. Pernet
age-related delay in
EEG sensitivity to
phase noise
Rousselet, Pernet et al.
BMC Neuroscience 2008
BMC Neuroscience 2009
Frontiers in Perception Science 2010
7
Joachim Gross
Research Interest:
Large-scale neural communication
Role of neural oscillations in cognition
Conscious perception
Methods:
Magnetoencephalography (MEG)
Spectral analysis
Granger causality
Graph Theory
Synchronisation analysis
8
cortical coding principles
8
i.e. cortical predictions Friston, K.
free energy principle
9
Lars Muckli – Predictive coding
9
10
Neurophysiological models for economic neural encoding
• Mumford 1992: On the computational
architecture of the neocortex
Attneave1954:“It appears likely that a major function of the perceptual machinery is
to strip away some of the redundancy of stimulation, to describe or
encode incoming information in a form more economical than that in
which it impinges on the receptors”
architecture of the neocortex
– Higher cortical areas try to predict
information that is present in lower
cortical areas based on abstract
information about the world.
– Lower areas signal only the information
that they contain which is not predicted
by higher areas.
– Predictions and error signals are
transmitted via a thalamo-cortical loop
11
Extension: V5 is sending a
spatial-temporal precise prediction to V1
Out-time In-time
*
*
Alink, Schwiedrzik,Kohler,Singer & Muckli (2010) J Neurosci.
12
Lars Muckli – Predictive Coding in Vision
1. ongoing activity before stimulus onset..
2. non stimulated regions
• during motion illusion
• during scene processing
V3
V1V2
13
spatial-temporal precise prediction in V1
0.5
In Time Flash
Out of Time FlashOut-time In-time
Stimulus
0 4 8 12 16
- 0.2
0
P < 0.00009
Alink, Schwiedrzik,Kohler,Singer & Muckli (2010) J Neurosci.
Reduced V1 BOLD in response to anticipated stimuli
Flash positionLower position
14
Our experiment, Part 1
Region-Of-Interest based Analysis Results:
Primary visual cortex 12 subjects
15
Disruption of feedback from V5 to V1 with TMS
Time (ms)
t1TMS TMS
t2TMS TMS
t3TMS TMS
t4TMS TMS
Vetter, Grosbras , & Muckli (in prep.)
0-13-53 +27 +107+67
Target onset
Time (ms)
** *n.s.
n.s.
16
The more predictable a visual stimulus is the less early visual areas will respond to that stimulus
V1 response amplitude
Our experiment, Part 2
Parametric Design:
Moving dots on the AM path with
a variable angle
V1 response amplitude
Unpredictability of stimulus
17
Our experiment, Part 2
18
Apparent motion in natural visual scenes
100 ms
800 ms
match
mismatch
ExpectancyExpectancy
match
…1 52 6 4 3
100 ms
800 ms
mismatch
Non Non -- ExpectancyExpectancy
Carvalho, Smith & Muckli (2009) HBM Meeting, in prep
19
Visual Stimulation
Carvalho, Smith & Muckli (2009) HBM Abstract
20
20
21
Predictive coding –
default mode network
21
Exp MatchExp non-match
Non-exp ‘match’
Non-exp ‘non-match’
22
Processing of Context Information AND AgingProcessing of Context Information AND Aging(Paxton et al 2006 Psychology and Aging)
Braver et al 2001
predictive coding
In DMN affected
by aging
22Paxton et al (2006)Psychology and Aging
23
Context decoding in V1 (brain reading 1st)
Smith & Muckli (under review)
24
25
Clinical applications
Models
Basic Research
Unifying theories of the brain
Free energy principle
Clinical applicationsPredictive Coding
Apparent motion
Complex scenes
Contextual information
Aging
Mental Disorder
AI applicationsApplications for artificial intelligence
Face recognition
Body motion recognition
26
..unifying theories of the brain
26
Friston et al., 2006: A free energy principle for the brainNeural systems attempt to minimize state changes by changing it’s
sampling of the environment.
This is realized by anticipating upcoming events based on learned
rules and the statistics of the environment
27
HTM Bayesian believe propagation (Markov chains)
27George & Hawkins (2010)
28
thanks to …
Dr Fraser W Dr Petra Vetter
28
MPI Frankfurt
– Arjen Alink
– Dr Axel Kohler
– Caspar Schwiedrzik
– Prof Wolf Singer
– Dr Michael Wibral
Dr. Fabiana
Carvalho
(Scenes)
Dr Fraser W
Smith
MVA-Faces-Scenes
Dr Petra Vetter
(TMS-saccades)