Download Autism and Computational Simulations

Autism
& computational simulations
Włodzisław Duch
Katedra Informatyki Stosowanej,
Uniwersytet Mikołaja Kopernika, Toruń, Poland.
Google: W. Duch
UoM KL workshop 2009
Plan:
How to understand autism and many other developmental, neurological and psychiatric problems?
• Computer Simulations of Brain Functions: general overview
of models, what do we want/can learn from them,
comparison of suitable neural simulators.
• Introduction to the Emergent simulator: general principles,
biological plausibility, learning algorithms, models.
• Understanding neural activity: visualization, recurrence
plots, fuzzy symbolic dynamics.
• Two models relevant to autism: attention shifts in visual
recognition; learning words.
• Conclusion: generative psychiatry.
Neurocognitive informatics
Computational Intelligence. An International Journal (1984)
+ 10 other journals with “Computational Intelligence”,
D. Poole, A. Mackworth R. Goebel,
Computational Intelligence - A Logical Approach.
(OUP 1998), GOFAI book, logic and reasoning.
• CI: lower cognitive functions, perception, signal analysis,
action control, sensorimotor behavior.
• AI: higher cognitive functions, thinking, reasoning, planning etc.
• Neurocognitive informatics: brain processes can be a great inspiration for AI
algorithms, if we could only understand them ….
What are the neurons doing? Perceptrons, basic units in multilayer perceptron
networks, use threshold logic – Artificial NN inspirations.
What are the networks doing? Specific sensory/motor system transformations,
implementing various types of memory, estimating similarity.
How do higher cognitive functions map to the brain activity? Still hard but …
Neurocognitive informatics = abstractions of this process .
Model of self-organization
SOMF, Self-Organized Feature Mapping, or Kohonen map.
Simplest model of topographic self-organization via competitive Hebbian
activity-dependent learning.
Signal X activates most
strongly a neuron with
synapses W; they become
more similar to X and also
neurons in the vicinity of W
become more similar to X.
Receptive fields of neurons
that are close on the 2D
map are close in the input
space.
Update equation:
Wi  t  1  Wi  t   h  ri , rc ,t   X  t   Wi  t   dla i  O  c 
Transformation model
Feedforward information processing, no recurrence:
categorization, info compression, sensomotoric action.
~ p(MI|X) 0.7 Myocardial Infarction
Output
weights
Input
weights
Inputs:
1
65
Sex
Age
1
5
3
1
Smoking Pain
Elevation
Pain
Intensity Duration ECG: ST
Dynamical Model
Strong feedback interactions, neurodynamics, collective states, recurrent
networks. Simplest model that serves as associative memory (Hopfield
1982): two-state neurons (active/inactive), Hebb learning rule,
asynchronic dynamics; replaced by graded neuron model.
Vector of input activations
V(0)=Vini , input = output = activations.
Discrete dynamics (iterations) 
Hopfield network may reach attractor,
interpreted as memory state, or
autoassociative response to the input
query Vini.
For symmetric connections (unrealistic)
this network reaches a stationary state
(point attractor).

t = quantized time.
Vi  t  1  sgn Ii  t  1  sgn 



j WijV j   j 


Biophysical model
Synapses
Soma
I syn ( t )
Spike
EPSP, IPSP
Rsyn
Spike
Csyn
Cm
,ext
I AMPA,ext (t )  g AMPA,ext (Vi (t )  VE ) wij s AMPA
(t )
j
j
,rec
I AMPA,rec (t )  g AMPA,rec (Vi (t )  VE ) wij s AMPA
(t )
j
j
I NMDA,rec (t ) 
g NMDA,rec (Vi (t )  VE )

(1  [ Mg 2 ]exp( 0.062Vi (t ) / 3.57)) j
,rec
IGABA,rec (t )  gGABA,rec (Vi (t )  VE ) wij s GABA
(t )
j
j
Rm
s AMPA
(t )
d AMPA
s j (t )   j
   (t  t kj )
dt
 AMPA
k
s NMDA
(t )
d NMDA
s j (t )   j
  x j (t )(1  s NMDA
(t ))
j
dt
 NMDA,decay
x NMDA
(t )
d NMDA
j
x j (t )  
   (t  t kj )
NMDA,rec
wij s j
(t ) dt
 NMDA,rise k
sGABA
(t )
d GABA
s j (t )   j
   (t  t kj )
dt
 GABA
k
Books & references
•
•
•
•
•
•
•
•
Reggia J.A, Ruppin E. and Berndt R.S, eds. (1996) Neural Modeling of Brain
and Cognitive Disorders. World Scientific.
Parks R.W, Levine D.S. and Long D, Eds. (1998) Fundamentals of Neural
Network Modeling. MIT Press, Cambridge, MA.
Reggia J.A, Ruppin E. and Glanzman D.L., Eds. (1999) Disorders of Brain,
Behavior, and Cognition: The Neurocomputational Perspective. Elsevier, NY.
O'Reilly R. & Munakata Y. (2000) Computational Explorations in Cognitive
Neuroscience: Understanding the Mind by Simulating the Brain. MIT Press.
Callaway E, Halliday R, Naylor H, Yano L, Herzig K. (1994) Drugs and human
information processing. Neuropsychopharmacology 10, 9-19.
Ruppin, E. (1996). Neural Modeling of Psychiatric Disorders. Network 6: 635656 (long review of early approaches).
Carnevale N.T, Hines M.L. (2006) The NEURON Book: Cambridge Uni Press.
Beeman D, GENESIS tutorial (2008)
http://www.genesis-sim.org/GENESIS/cnslecs/cnslecs.html
Brain-like computing
Brain states are physical, spatio-temporal states of neural tissue.
• We see, hear and feel brain states, mostly internal dynamics !
• Cognitive processes operate on highly processed sensory data.
• Redness, sweetness, itching, pain ... are all physical states of brain tissue.
Ex: visual illusions.
In contrast to computer registers,
brain states are dynamical, and thus
contain in themselves many
associations, relations.
Inner world is real! Mind is based
on relations of brain’s states,
concepts = pdf(brain activations).
Computers and robots do not have
an equivalent of such WM.
Symbols in the brain
Organization of the word recognition circuits in the left temporal lobe
has been elucidated using fMRI experiments (Cohen et al. 2004).
How do words that we hear, see or are thinking of, activate the brain?
Seeing words: orthography, phonology, articulation, semantics.
Lateral inferotemporal multimodal area (LIMA) reacts to auditory visual
stimulation, has cross-modal phonemic and lexical links. Adjacent visual word
form area (VWFA) in the left occipitotemporal sulcus is unimodal.
Likely: homolog of the VWFA in the auditory stream, the auditory word form area,
located in the left anterior superior temporal sulcus.
Large variability in location of these regions in individual brains.
Left hemisphere: precise representations of symbols, including phonological
components; right hemisphere? Sees clusters of concepts.
Words in the brain
Psycholinguistic experiments show that most likely categorical,
phonological representations are used, not the acoustic input.
Acoustic signal => phoneme => words => semantic concepts.
Phonological processing precedes semantic by 90 ms (from N200 ERPs).
F. Pulvermuller (2003) The Neuroscience of Language. On Brain Circuits of Words
and Serial Order. Cambridge University Press.
Action-perception
networks inferred
from ERP and fMRI
Phonological neighborhood density = the number of words that are similar in
sound to a target word. Similar = similar pattern of brain activations.
Semantic neighborhood density = the number of words that are similar in meaning
to a target word.
Neuroimaging words
Predicting Human Brain Activity Associated with the Meanings
of Nouns," T. M. Mitchell et al, Science, 320, 1191, May 30, 2008
•
•
•
•
Clear differences between fMRI brain activity when people read and think
about different nouns.
Reading words and seeing the drawing invokes similar brain activations,
presumably reflecting semantics of concepts.
Although individual variance is significant similar activations are found in brains
of different people, a classifier may still be trained on pooled data.
Model trained on ~10 fMRI scans + very large corpus (1012) predicts brain
activity for over 100 nouns for which fMRI has been done.
Overlaps between activation of the brain for different words may serve as
expansion coefficients for word-activation basis set.
In future: I may know what you’ll think before you will know it yourself!
Intentions may be known seconds before they become conscious!
Energies of trajectories
P. McLeod, T. Shallice, D.C. Plaut,
Attractor dynamics in word recognition: converging evidence from errors by
normal subjects, dyslexic patients and a connectionist model.
Cognition 74 (2000) 91-113.
M. Spivey, Continuity of mind. Oxford University Press 2007
New area in psycholinguistics: investigation of dynamical cognition, influence of
masking on semantic and phonological errors.
Problems requiring insights
Given 31 dominos
and a chessboard with 2 corners
removed, can you cover all board with dominos?
Analytical solution: try all combinations.
Does not work … to many combinations to try.
Logical, symbolic approach has
little chance to create proper
activations in the brain, linking
new ideas: otherwise there will
be too many associations,
making thinking difficult.
Insight <= right hemisphere,
meta-level representations
without phonological (symbolic)
components ... still important ,
subconscious thinking, insights.
chess board
domino
n
black
white
m
do
o
i
phonological reps
Insights and brains
Activity of the brain while solving problems that required insight and that
could be solved in schematic, sequential way has been investigated.
E.M. Bowden, M. Jung-Beeman, J. Fleck, J. Kounios, „New approaches to
demystifying insight”. Trends in Cognitive Science 2005.
After solving a problem presented in a verbal way subjects indicated themselves
whether they had an insight or not.
An increased activity of the right hemisphere anterior superior temporal
gyrus (RH-aSTG) was observed during initial solving efforts and insights.
About 300 ms before insight a burst of gamma activity was observed,
interpreted by the authors as „making connections across distantly related
information during comprehension ... that allow them to see connections
that previously eluded them”.
Insight interpreted
What really happens? My interpretation:
•
•
•
•
•
•
•
•
LH-STG represents concepts, S=Start, F=final
understanding, solving = transition, step by step, from S to F
if no connection (transition) is found this leads to an impasse;
RH-STG ‘sees’ LH activity on meta-level, clustering concepts into
abstract categories (cosets, or constrained sets);
connection between S to F is found in RH, leading to a feeling of vague
understanding;
gamma burst increases the activity of LH representations for S, F and
intermediate configurations; feeling of imminent solution arises;
stepwise transition between S and F is found;
finding solution is rewarded by emotions during Aha! experience;
they are necessary to increase plasticity and create permanent links.
Types of memory
Neurocognitive approach needs at least 4 types of memories.
Long term (LTM): recognition, semantic, episodic + working memory.
Input (text, speech) pre-processed using recognition memory model to correct
spelling errors, expand acronyms etc.
For dialogue/text understanding episodic memory models are needed.
Working memory: an active subset of semantic/episodic memory.
All 3 LTM are coupled mutually providing context for recognition.
Semantic memory is a permanent storage of conceptual data.
• “Permanent”: data is collected throughout the whole lifetime of the
system, old information is overridden/corrected by newer input.
• “Conceptual”: contains semantic relations between words and uses them
to create concept definitions.
Memory in Alzheimer
AD and various diseases leading to dementia are connected with decrease
of the density of weak synaptic connections.
What happens with associative memory in simple models if weak
connections are removed? Is there a biological mechanism that can help to
compensate for such memory loss?
Horn D, Levy N, Ruppin E (1996) Neuronal-based synaptic compensation: A
computational study in Alzheimer's disease. Neural Comput 8, 1227-1243.
Hopfield model shows how the remaining synapses may adapt to minimize
memory damage.
dk 

Wij  Wij 1 

 1 d 
n
o
d – degree of impairment
k=k(d) compensation function
Strong synapses should become even stronger.
The degree of memory impairment depends not only on d but also
on the function k(d) related to the history => various symptoms with
the same synaptic density are observed.
Compensation
The size of basins of
attractors as a function of the
percent of network damage.
Upper circle: learning
without compensation; lower
– with compensation.
Dominant attractors shrink
and small attractors have a
chance to be active.
Performance = % of correct
memory associations as a
function of synaptic deletion
parameter d without
compensation (dot-dash)
and with two forms of
compensation (dashed,
continuous lines).
Trace-Link model
Meeter M & Murre J.M.J. (2004) Simulating episodic memory deficits
in semantic dementia with the TraceLink model. Memory 12, 272-287
System 1: Trace system
• Function: Substrate for bulk storage of memories, ‘association
machine’
• Corresponds roughly to neocortex
Slides: courtesy of Jaap Murre
System 2: Link system
• Function: Initial ‘scaffold’ for episodes
• Corresponds roughly to hippocampus and certain temporal
and perhaps frontal areas
Slides: courtesy of Jaap Murre
System 3: Modulatory system
• Function: Control of plasticity
• Involves at least parts of the hippocampus,
amygdala, fornix, and certain nuclei in the basal
forebrain and in the brain stem
Slides: courtesy of Jaap Murre
Stages in episodic learning
Retrograde amnesia
• Primary cause: loss of links
• Ribot gradients
• Shrinkage
Anterograde amnesia
• Primary cause: loss of modulatory system
• Secondary cause: loss of links
• Preserved implicit
memory
Slides: courtesy of Jaap Murre
Semantic dementia
• The term was adopted recently to describe a new form of
dementia, notably by Julie Snowden et al. (1989, 1994) and by
John Hodges et al. (1992, 1994)
• Semantic dementia is almost a mirror-image of amnesia
• Progressive loss of semantic knowledge
• Word-finding problems
• Comprehension difficulties
• No problems with new learning
• Lesions mainly located in the infero-lateral temporal cortex
but (early in the disease) with sparing of the hippocampus
Slides: courtesy of Jaap Murre
Semantic dementia in TraceLink
• Primary cause: loss of trace-trace connections
• Stage-3 (and 4) memories cannot be formed: no consolidation
• The preservation of new memories will be dependent on
constant rehearsal
Slides: courtesy of Jaap Murre
No consolidation in semantic dementia
Severe loss of trace
connections
Stage 3 learning strongly
impaired
Stage-2 learning proceeds
as normal
Non-rehearsed memories
will be lost
Semantic Memory Models
Endel Tulving „Episodic and Semantic Memory” 1972.
Semantic memory refers to the memory of meanings and understandings.
It stores concept-based, generic, context-free knowledge.
Permanent container for general knowledge (facts, ideas, words etc).
Hierarchical Model
Collins Quillian, 1969
Semantic network
Collins Loftus, 1975
SM & neural distances
Activations of groups of neurons presented in activation space define
similarity relations in geometrical model (McClleland, McNaughton, O’Reilly,
Why there are complementary learning systems, 1994).
Similarity between concepts
Left: MDS on vectors from neural network.
Right: MDS on data from psychological experiments with perceived similarity
between animals.
Vector and probabilistic models are approximations to this process.
Sij ~ (wi,Cont)|(wj,Cont) 
Neurological models
Epilepsy: infinite variations.
Positive feedback or lack of sufficiently strong inhibition is a general
metaphor, but biophysical models are more precise.
Detailed models of pyramidal neurons and interneurons in the CA3 area of
hippocampus elucidated synchronization processes and showed the
influence of various chemicals.
Very high 200-600 Hz (phi) frequencies observed in some form of epilepsy
cannot be generated by “normal” chemical synapses.
Fast electrical nonsynaptic communication is possible through gap junctions
filled with connexins, intramembranous proteins, that have rapidly
modifiable conductance properties.
At least two such synapses/neuron are needed (Traub et al, Nature 1996).
How to block synchronization leading to epilepsy?
It will require understanding of processes at the molecular level.
Stroke and cortical reorganization
Topographical representations are not fixed, they may reorganize as a result
of lesion due to stroke, nerve damage or limb amputation (no activation).
Stimulation of adjacent cortical areas may lead to quick expansion of
neural representations in S1 cortex to unused areas that will now react to
stimulation of quite different parts of skin.
Simple SOM models give qualitatively correct results, but transmission
through the thalamus is needed to achieve „reverse magnification” effect,
size of S1 representation is inversely proportional to the size of receptive
field.
Phantom limbs: detailed models are challenging
because large (10 mm) reorganization of
thalamic projections are observed while usually
reorganization is limited to ~ 1 mm.
This may indicate that quite new connections
are developed.
Simulation of fast reorganization
M. Mazza et al, J. Computational Neuroscience 16 (2004) 177-201, detailed
model implemented in GENESIS.
Hand: 512=32*16 receptors, palm 4x32
+ 4 fingers 8x12, Meissner corpuscles
mechano-receptors sending bursts of action
potentials to ventral posterior lateral (VPL)
part of the thalamus.
AMPA, GABA & NMDA synaptic receptor
dynamics was approx. with simple functions.
VPL: thalamic relay cells and interneurons.
Ion channels on the soma:
Na, Ca low threshold
inactivating Ca (Cat),
fast Ca voltage dependent K
(Kc), slow calciumdependent K (Kahp),
delayed rectifier K (Kd).
VPL Model
Grid 16x16 = 256 relay neurons
+ 128 interneurons.
Input from hand receptors
through AMPA synapses, from
other relay neurons through
AMPA and NMDA receptors.
Inhibition through GABA
receptors.
Connection probabilities
between neurons taken from
neuroanatomical data.
Each cell is connected inside the radius Rc with other cells with p=0.5,
in the ring Rc-Re, in elliptical area for the relay neuron with NMDA channels,
and interneurons with GABA receptors.
3b cortex area
Only layers II, IV i V are
modeled, each 32x32=1024.
Neuron types:
A: burst spiking neurons(BSN):
excitatory stellate cells layer IV,
B: fast spiking inhibitory
GABAergic basket cells (FSN).
C: excitatory pyramidal V,
D: excitatory pyramidal layer III,
regular spiking neurons (RSN);
Pyramidal neurons (C, D) have 8
segments, stellate (A) 5, a total of 3072
excitatory neurons in the network.
Interneurons (B) have 2 segments, a
total of 1536, with 512 in each layer.
3b connectivity
A, stellate neurons of layer
IV, input from thalamus and
other neurons in layer IV.
B: basket cells connected to
various excitatory cells.
C & D: excitatory pyramidal
layer III + V cells connected
to thalamus and layers II, IV
and V.
Connections are defined
according to experimentally
derived probabilities.
Equations for currents/potentials are integrated with time step 0.05 ms, on a
slow PC 1 sec. of real time required 3.5 hour of CPU time.
This model allows for creation of topographical maps in VPL nuclei and in the
3b layers of neocortex.
Simulation results
Simulation of randomly selected small 2x2 patches of
hand receptors, in 1 s sending 20.000 activations.
VPL activity becomes stable already after 500 ms, and
cortical activity in layer II and V after 750 ms, and in layer
IV after 800 ms.
Light areas = weak mixed response, dark >10 spikes/s
Results:
 Representation area of palm is smaller than fingers.
 Layer IV has developed most precise map (this
agrees with experimental data).
 Fluctuations are constantly present on the borders that
are “dynamically maintained”.
Amputation of finger 2
After stability is reached (900 ms) and the map formed all
stimuli coming from finger 2 were removed.
After 400 ms the map has reorganized itself.
Some part of neurons stopped reacting to activations
but representation of finger 1 and 3 have increased in
size, especially in the layer II and V, to a smaller degree in
the layer IV and in the thalamus.
Simulations demonstrate rapid expansion and
reorganization of cortical representations, after that
slower consolidation of changes follows.
Stability of maps is a result of balance between excitation
and inhibition; reorganization results from decrease of
inhibition and increase of activity of NMDA receptors.
LTP plasticity has not been included in this model.
Emergent
Emergent
Comparison of neural simulators.
Emergent: a comprehensive simulation environment for creating complex,
sophisticated models of the brain and cognitive processes using neural
network models. Full 3D GUI environment for constructing networks and the
input/output patterns for the networks to process, and many different
analysis tools for understanding what the networks are doing.
Aisa, B., Mingus, B., and O'Reilly, R. (2008). The emergent neural modeling
system. Neural Networks, 21, 1045-1212.
Randall C. O'Reilly and Yuko Munakata (2000) Computational Explorations in
Cognitive Neuroscience Understanding the Mind by Simulating the Brain,
Cambridge, MA: MIT Press.
Started as PDP++ Neural Network Simulator (1993).
In 2009 Emergent 5.0 was introduced.
http://grey.colorado.edu/emergent/index.php/Main_Page
Many project were done in Emergent, including SAL (new BICA architecture),
some project require older version of software (4.0.19 and PDP ++).
Some things you can do with Emergent
Many tutorials and projects are described in the book and are ready for
explorations, including:
1. Development of visual cortex receptive fields, properties of these fields.
Why does primary visual cortex encode oriented bars of light?
2. How is objects recognition invariant across locations, sizes, rotations
possible? Series of transformations from V1=>v2=>V4=>IT
3. How is visual attention realized? Why is visual system split into
what/where pathways? How does V2=>PPC=>V4 help to shift attention?
4. Why does parietal damage cause attention problems (unilateral neglect)?
5. Memory: how do our brains find the balance between the need to
associate and to remember the details? How do we learn AB-AC lists
without catastrophic forgetting? How do working, episodic and semantic
memory differ and interact?
6. How are higher cognitive functions realized? Reading, dyslexia, learning
irregular verbs, learning meaning, categorization …
Emergent: wyniki
Neurons in Emergent
Vm is equilibrium membrane potential, <xiwij> means time average, Q is
threshold for activity, b = const for different types of neurons (slow, fast).
Output: spikes or rate coding,
number of spikes per second.
[ . ]+ positive or 0.
ge, gi, gl are synaptic
conductance values.
Activation for rate coding is of the
gx/(gx+1)=x/(x+1/g) type; gaussian
noise is added to smooth this function,
contributing to its sigmoidal shape,
parametrer g regulates its slope.
Spikes
Simplified model that includes 3 types of channels: excitatory, inhibitory
and leak channels.
Emergent: neuron simulation
Net = total activation changing
from 0 to g_bar_e=1 when all
channels are opened.
I_net: current flows, equilibrium is
reached and sharply drops to zero,
after net=0 some current flows
back.
V_m is the potential on the axon
hillock, growing from -70mV (here
0.15) to +50mV (here 0.30).
Act = activity along the axon;
if spike coding is used single
pulses are sent, some noise
(here with 0.001 variance) may create small fluctuations.
Act eq = rate-code, total cummulative average activity.
Leak channels (K+)
Change of conductivity of the leak channels has an influence on selectivity of
neurons (size of basins of attractors), for large ĝl only one unit reacts,
for small ĝl more units are involved and variability is larger.
More associations  lower precision of recognition.
ĝl =6
ĝl = 5
ĝl = 4
NMDA receptors
1. Mg+ ions block NMDA channels.
When the postsynaptic potential grows
these ions are released and glutamate
may bind to the channel.
2. Glutamate is released by presynaptic
spikes and is ready to open NMDA
channels.
3. Ca++ ions enter this channel initiating a
number of chemical reactions in the
neuron.
Effects are nonlinear: small amount of Ca++
lead to LTD, large to LTP.
Many other processes play a role in LTP.
kWTA approximation
The k Winners Take All (kWTA) is frequently used approximation to leave only
small number k of active neurons; this is simpler and faster than using direct
inhibition by interneurons, reaching similar sparse representations.
Inhibitory neurons reduce activity, the idea is to keep the system around the
point where no more than k neurons are simultaneously active.
1. Find k most active neurons in a given layer.
2. Calculate appropriate inhibition to leave only k above the firing threshold.
Distribution of neuron activity level in a
larger network should have
approximately Gaussian character.
A threshold level should be defined
such that giQ should balance this
excitations for between k and k+1
neurons.
Connectivity of networks
1.
2.
3.
Input layer 4, initial processing of external (sensory) information.
Hidden layers 2/3, further processing of inputs, associations, mostly internal
activations.
Output layers 5/6, subcortical connections and motor system.
Neurons and detectors
Neurons working as simple detectors
are used in pandemonium:
Selfridge has proposed already in
1959 distributed architecture in:
Symposium on the mechanization of
thought processes.
London: HM Stationary Office.
Pandemonium was used for character recognition and was quite popular in
cognitive psychology.
Demons watch incoming data, trying to recognize familiar traits, if they are
successful they shout, the more confident they are the louder they shout.
Second row demons listen to a few selected ones from the first row and shout
further if they hear sufficiently high activity.
Final decision is taken by the high-level demon.
This is essentially the multi-layer perceptron (MLP) feedforward architecture.
Pandemonium in action
Demons observing fetures:
| vertical segment
D1
-- horizontal segment
D2
/ right-slanted segment D3
\ left-slanted segment D4
V
T
A
K
demons 3, 4
demons 1, 2
demons 2, 3, 4
demons 1, 3, 4
=> D5
=> D6
=> D7
=> D8
Better fit = lauder shouts.
Decision demon D9 here does not distinguish CAT and ACT ...
Although all demons take very simple decisions the whole may realize
functions of arbitrary complexity.
Hebb rule and correlations
Networks should reflect (internalize) properties of environment , correlations
between sensory signals.
Simple Hebb rule: Dwij = e ai aj
The change of synaptic weights is proportional to pre & post-synaptic
potentials.
Weights are increasing for neurons that have correlated activity, but are not
changed for neurons that do not show any correlations.
Normalized Hebb rule
Simple Hebb rule:
Dwij = e xi yj
leads to an infinite growth of weights.
The simplest way to avoid it is to normalize weights:
Dwij = e (xi -wij) yj
x is presynaptic signal, y postsynaptic signal.
Biological justification:
• when x and y are large than LTP, high concentration of Ca++
• when y is large but x is small then LTD, small concentration of Ca++
• when y is small nothing happens because Mg+ blocks NMDA channels.
GeneRec learning
Most connectionist models in cognitive psychology is based on
backpropagation algorithms applied to multilayer perceptrons (that can learn
an arbitrary mapping), but this algorithm is hard to justify from biological
point of view. O’Reilly (1996) introduced a two-phase algorithm:
GeneRec (General Recirculation).
Signals are propagated both ways,
weights are non-symmetric
wkl  wjk.
First the – phase, or network
response to activation x– = si
gives outputs y– = ok, then external
target y+ = tk is propagated towards
inputs layer x+.
Change of weights is based on
information from both phases.
Learning with GeneRec
Learning rule is based on modification of Hebb rule (also called delta rule):
Dwij  e  y j  y j  xi
The difference between the signals is an error signal [y+  y].
For biases b inputs are xi=1, so:


Db j  e  y j  y j 
Gradient rule: signal difference is ~ activation difference * derivative of
activation function (linear Taylor expansion).
Bidirectional information passing is very quick (P300 signal shows the
expectations) and is responsible for formation of top-down/bottom-up
recurrent loops leading to expectation-based pattern completion.
Errors are result of the whole network activity, slightly faster learning is
obtained taking average [x++x-]/2 and using symmetric weights:
Dwij  e  xi y j  xi y j 
Contrastive Hebbian Rule,
or CHR.
Target signals
Where does the target for corrections comes from?
Output on the right side at t+1 becomes middle at t+0.5.
Pronunciation of words a) & c) external corrections of actions;
b) implicit expectation; d) memory-based reconstruction of action.
GeneRec with Hebbian learning
Hebbian learning creates a model of the world, remembers correlation but is
not able to learn tasks that require mapping from input to output.
Hidden layers allow to transform problem extracting relevant features while
correction of errors enables learning of difficult input/output tasks.
Combination of Hebbian correlation learning and error correction may learn
and task in biologically plausible way: with bidirectional corrections CHL leads
to approximate symmetry, Err = CHL in the weight change table below.
No Ca2+ = no learning; little Ca2+ = LTD, high Ca2+ = LTP
LTD – incorrect expectations, only the – phase, but no confirmation in the +
phase (failed expectations lead to depression).
Full Leabra model
6 principles of biologically inspired neural network.
Inhibition inside the layers, Hebbian learning + error-driven correction to
train connections between the layers.
3 components of BICA
PC – posterior parietal cortex and some
part of motor cortex; sensomotoric
actions, associations, distributed reps.
FC – prefrontal cortex, higher cognitive
functions, working memory, isolated reps.
HC – hippocampal formation, episodic and
spatial memory, fast learning, sparse reps.
• Learning must be slow to capture statistical correlations and analyse precisely
data from sensory systems, control motor system.
Fast learning is also important.
• Solution: slow semantic cortical learning (PC) and fast episodic learning (HC).
• Keep active information and at the same time analyze new information in a
distributed system, avoid interference.
Biologically Inspired Cognitive Architecture
BICA: hierarchical structure for sensory data in parietal cortex (PC),
adding context to episodes in hippocampus (HC),
Localized, recurrent competing representations in frontal cortex (FC).
Autism
Symptoms ...
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Difficulty in mixing with other children.
Prefers to be alone; aloof manner.
Inappropriate laughing and giggling.
Inappropriate attachment to objects.
Little or no eye contact.
May not want cuddling or act cuddly.
Apparent insensitivity to pain.
Spins objects; sustained odd play.
Insistence on sameness; resists changes in routine.
Noticeable physical overactivity or extreme underactivity.
Unresponsive to normal teaching methods.
No real fear of dangers.
Echolalia (repeating words or phrases in place of normal language).
Not responsive to verbal cues; acts as deaf.
Difficulty in expressing needs; uses gestures or pointing instead of words.
Tantrums - displays extreme distress for no apparent reason.
Uneven gross/fine motor skills (no kicking of balls but can stack blocks).
Symptoms ...
Pathophysiology
• Alteration of brain development soon after conception, significantly
influenced by environmental factors.
• Is there a unifying mechanism at molecular, cellular, or systems level?
• Autism may result from a few disorders caused by mutations converging on
a few common molecular pathways.
• Autism may be a large set of disorders with diverse mechanisms, like
intellectual disability.
• An excess of neurons that causes local overconnectivity in key brain regions.
• Disturbed neuronal migration during early gestation.
• Unbalanced excitatory–inhibitory networks.
• Abnormal formation of synapses and dendritic spines, poorly regulated
synthesis of synaptic protein, also associated with epilepsy.
Geschwind DH (2008). Autism: many genes, common pathways? Cell 135: 391–5;
Müller RA (2007) The study of autism as a distributed disorder. Mental
Retardation and Developmental Disabilities Research Reviews 13 (1): 85–95
Casanova MF (2007) The neuropathology of autism. Brain Pathology 17: 422–33
More pathophysiology
• Autism is a multifactorial or complex trait.
• Genetic studies report that monozygotic twins have a concordance rate of
75% as opposed to 3% for fraternal twins. Families having an autistic member
have a 10–40% increased incidence of other developmental disorders.
• Many different genes may be involved in the expression of autism. In 2009
PTEN and the serotonin transporter gene were implicated in large brain size
and poor social behavior in mice.
• Genes act in an additive way along with the environment to produce the final
phenotype. Underlying pathologies to a multifactorial trait exhibit a
continuous distribution of changes.
• ERP studies: differences in autistic individuals with respect to attention,
orientation to auditory and visual stimuli, novelty detection, language and
face processing, and information storage; several studies have found a
preference for non-social stimuli.
• MEG studies: delayed responses in the brain's processing of auditory signals.
Neuroanatomy
• Increased brain size in childhood is the most robust macroscopic feature of
autism, the difference (10-20%) disappears with age.
• Increased cerebral gray and white matter and cerebellum.
• Most significant is the frontal gray and white matter intrahemispheric
volume increase occurring in the first 2-4 years.
• Widespread cortical abnormalities, disruption of laminar organization and
heterotopias (displacement of gray matter into white matter or ventricles).
• Mental retardation in 60–70% of cases.
• Absence of spasticity or vision/hearing loss.
• Seizures in about 30% of cases.
• 40% autistic children have some form of epilepsy.
• No focal dysfunctions, distributed neocortical system disorder.
Common belief: more frequently problems with association/prefrontal cortex
than sensory/motor areas.
Understanding
Modular (bottom-up) and system biology approaches (interacting levels),
M.R. Herbert, M.P. Anderson, Chap 20, Zimmerman.
Causality is not linear, top-down and bottom up influences are possible.
Theories, theories
Best book so far:
• Zimmerman Andrew W. (Ed.)
Autism; current theories and evidence.
Humana Press 2008.
• 20 chapters divided into six sections:
• Molecular and Clinical Genetics (4 chapters);
• Neurotransmitters and Cell Signaling (3 chapters);
• Endocrinology, Growth, and Metabolism (4 chapters);
• Immunology, Maternal-Fetal Effects, and Neuroinflammation (4 chapters);
• Neuroanatomy, Imaging, and Neural networks (3 chapters);
• Environmental Mechanisms and Models (2 chapters).
Mirror Neuron System
• The mirror neuron system (MNS): multimodal neurons, in motor cortex,
react also to visual observations, observing action elicits similar motor
activations as if it had been performed by oneself.
• The MNS helps to understand actions of others, modeling their behavior via
embodied simulation of their actions, intentions, and emotions.
• MNS theory of autism: distortion in the development of the MNS interferes
with the ability to imitate, leads to social impairment and communication
difficulties.
• Structural abnormalities in MNS regions of individuals with ASD exist,
correlations between reduced MNS activity and severity of ASD.
• But … in ASD abnormal brain activation in many other circuits; performance
of autistic children on various imitation tasks may be normal.
• MNS is not really a special system … the idea is used to explain almost
everything in social neuroscience.
MNS EEG
• EEG on controls and autistics on 4 different tasks, comparing mu rhythms.
At baseline, large amplitude mu oscillations in synchrony. Seeing an action
causes mu rhythms to fire asynchronously resulting in mu suppression.
• So mu wave suppression will reflect activity of the mirror neuron system.
• In autistics mu is suppressed for own hand movements, but not for the
observed hand movements of others.
Reduced functional connectivity
The underconnectivity theory of autism is based on the following:
• Excess of low-level (sensory) processes.
• Underfunctioning of high-level neural connections and synchronization,
• fMRI and EEG study suggests that adults with ASD have local
overconnectivity in the cortex and weak functional connections between
the frontal lobe and the rest of the cortex.
• Underconnectivity is mainly within each hemisphere of the cortex and
that autism is a disorder of the association cortex.
• Patterns of low function and aberrant activation in the brain differ
depending on whether the brain is doing social or nonsocial tasks.
• “Default brain network” involves a large-scale brain network (cingulate
cortex, mPFC, lateral PC), shows low activity for goal-related actions; it is
active in social and emotional processing, mindwandering, daydreaming.
• Activity of the default network is negatively correlated with the “action
network” (conscious goal-directed thinking), but this is not the case in
autism – perhaps disturbance of self-referential thought?
Empathizing–systemizing theory
• The extreme male brain theory: autism as an extreme case of the male
brain, those individuals in whom systemizing is better than empathizing
(according to psychometrical tests).
• Systemize = develop internal rules to handle events inside the brain.
• Empathize = rules handling events generated by other agents.
• Explains why more boys have autism, but baby boys and girls do not
respond differently to people and objects.
• Theory of mind: autism arises from inability to ascribe mental states to
oneself and others, as shows in the results of tests for reasoning about
others' motivations.
• Agrees with the mirror neuron system theory of autism.
• Many aspects are not addressed, very superficial understanding …
S. Baron-Cohen, Autism: the empathizing-systemizing (E-S) theory.
Ann N Y Acad Sci. 1156:68-80, March 2009.
Executive dysfunction
• Executive dysfunction hypothesis: autism results mainly from deficits
in working memory, planning, inhibition, and other executive functions.
• Executive processes such as voluntary eye movements slowly improve in
time but do not reach typical adult levels.
• Predicts stereotyped behavior and narrow interests.
• No executive function deficits have been found in young autistic children.
• Weak central coherence theory hypothesizes that a limited ability to see
the big picture underlies the central disturbance in autism.
• One strength of this theory is predicting special talents and peaks in
performance in autistic people.
• Enhanced perceptual functioning theory focuses more on the superiority
of locally oriented and perceptual operations in autistic individuals.
• These theories map well from the underconnectivity theory of autism.
• Social cognition theories poorly address autism's rigid and repetitive
behaviors, while the nonsocial theories have difficulty explaining social
impairment and communication difficulties.
Function connectivity theory
Model developed over 20 years (Nancy J. Minshew): autism as widespread
disorder of association cortex, development of connectivity, only secondarily
as a behavioral disorder. Fine, but still quite general.
Abnormalities in genetic code for brain development

Abnormal mechanisms of brain development

Structural and functional abnormalities of brain

Cognitive and neurologic abnormalities

Behavioral syndrome
Goal: understand the pathophysiology from gene to behavior, eventually the
influence of etiologies on this sequence, ultimately support the development
of interventions at multiple levels of the pathophysiologic sequence.
FC many names
Minshew model of autism: “complex information processing disorder”,
“connectivity/disconnectivity/ underconnectivity disorder”, a “disorder of
cortical development”, a “neuronal organization disorder”, intact or enhanced
simple information processing, but poor complex/higher order processing.
Integration of multiple features, processing of large amounts of information,
or novel material requires association cortex, but not sensory or motor
cortices – no blindness or deafness => problems mostly in prefrontal areas?
Preschool ASD children repeat words without comprehending, and/or
spontaneously use those words in an original way.
Most severe ASD: little to no development of functional connections between
sensorimotor cortex and association cortex, and thus no meaning was being
attached to information => stronger impact on connectivity with frontal
cortex than connectivity with unimodal cortex; reduced connections between
the memory and executive systems … common denominator is a dependence
on the degree of integration at the information processing level.
But why these connections are so weak?
Grossberg iSTART
iSTART, Imbalanced Spectrally Timed Adaptive Resonance Theory (2004).
START model developed by Grossberg to explain how the brain controls normal
behaviors, based on his ART (Adaptive Resonance Theory) theory.
Interactions of cognitive, emotional, timing, and motor processes involving
prefrontal and temporal cortex, amygdala, hippocampus, and cerebellum
create autistic symptoms.
Breakdowns in these brain processes:
• under-aroused emotional depression in the amygdala/related brain regions,
• learning of hyperspecific categories in temporal and prefrontal cortices,
• breakdown of adaptively timed attentional and motor circuits in the
hippocampal system and cerebellum.
Malfunctions in a subset of these mechanisms through a system-wide vicious
circle of environmentally mediated feedback cause and maintain problems.
Interesting but complex, hard to connect to molecular level.
iSTART specifics
Autistic people have vigilance fixed at such a high setting that their learned
representations are very concrete, or hyperspecific, which perpetuates a
multitude of problems with learning, cognition, and attention.
Cognitive-Emotional-Motor (CogEM), model, extends ART to the learning of
cognitive-emotional associations between events and emotions that give
these events value; under- or over-arousal can cause abnormal emotional
reactions and problems with cognitive-emotional learning.
If the emotional circuits are under-aroused, the threshold for activating an
emotion is abnormally high, but when this threshold is exceeded, the
emotional response can be over reactive - individuals with autism experience
reduced emotional expression as well as emotional outbursts.
Spectral Timing model: failures of adaptive timing that lead to the premature
release of behaviors which are then unrewarded.
Mapping to brain regions is possible - hyperspecific recognition: temporal and
prefrontal cortices, emotions: amygdala, timed attention and motor circuits in
the hippocampal system and cerebellum.
Autism: computational models
Theories developed so far are too general.
Ideal model should:
1.
2.
3.
Connect neural properties with behavior.
Connect neural properties with molecular level.
Connect molecular abnormalities with genetic level.
Strategy for the first step:
• Create models of various cognitive functions that are impaired in a given
disease; start with the most basic functions.
• Investigate influence of neural properties on these functions.
• Try to see how such impaired networks may create problems with other
cognitive and affective functions.
In case of autism the most basic deficit seems to be the inability to shift
attention – how do we shift attention and what may go wrong?
Excitatory and inhibitory neurons
Glutamic acid/ACh
opens Na+ excitatory
channels.
GABA inhibits neural
activity working on Clchannels.
Accommodation
A long depolarizing pulse => a single action potential.
Depolarization turns on a slowly activated K+ current (M-type channels).
Slow activation kinetics => one action potential before enough K+ goes out to
decrease membrane potential preventing firing. Acetylcholine (ACh) closes
the M-type channels, increasing cell activity (Jones & Adams, 1987).
M-type channels control excitability.
Retina
• Retina is not a CCD that passively registers images.
• Key principle: find contrasts in space/time, discover edges,
uniform surfaces are not so important.
• Fotoreceptors in cones (7M) and rods (100M).
• 3-layered network, ganglion cells=>1M fibers LGN.
Receptive field: the area that
strongly activates ganglion cell.
Over 100M receptors reduce info to
1M transmission lines providing oncenter) and off-center receptive
fields, extracted from receptor
signals by bipolar + ganglion cells.
This enhances the edges.
Such information arrives in LGN
and than visual cortex with a speed
estimated at 9 Mbps.
Image in V1
• Model is mostly focused on edge detectors, as this is the most important
function of primary visual area.
Vision
• From retina through lateral geniculate body, LGN (part of thalamus)
information passes to the primary visual cortex V1 and then splits into the
ventral and dorsal streams.
Recognition of many objects
• Vision model including LGN, V1, V2, V4/IT, V5/MT
Two objects are presented.
Connectivity of these layers:
Spat1  V2, Spat 2
Spat1  V2, Spat 2
Spat2  V2.
Spat1 has recurrent
activations and inhibition,
focusing on a single object.
In normal situations neurons
desynchronize and
synchronize on the second
object = attention shift.
Visualization: motivation
•
•
•
•
•
•
•
Analysis of multi-channel, non-stationary, time series data x(t).
Neural models: layers with hundreds of neurons quickly changing their
activity, what is the system doing as a whole?
Information is in the trajectories, but most methods focus on component
analysis and transformations (ICA); how to see trajectories? So far:
Component-based analysis: find independent components.
Time-frequency analysis: find where the energy is at each moment in time.
Recurrence plots: when x(t) comes close to the same point?
Symbolic dynamics: which box is the trajectory in?
Fuzzy Symbolic Dynamics (FSD), visualize + understand.
1. Reduce dimensionality of x(t), visualize in 2D or 3D.
2. Estimate the number of attractors and their properties.
3. Combined with time-frequency or component analysis.
Single neural activity
Too many channels, hard to analyze.
Neural activity
Event related synchronization (red) and desynchronization in time-frequency
plots.
Global visualization
Trajectory of dynamical system (neural activities, av. rates):
1.. N
{xi (t )}it 1..
n
Put localized probe function in Rn space:
sharp indicator functions => symbolic dynamics: ABBCAABB …
small neighborhoods => recurrence plots, Rk (x,e)=Q(e||x-xk||)
soft membership functions => fuzzy symbolic dynamics: (y1,y2,y3).
1. Standardize data.
2. Find cluster centers (e.g. by k-means algorithm): m1,
3. Use non-linear mapping to reduce dimensionality:

m2 ...
yk (t; mk , k )  exp   x  mk  k 1  x  mk 
T

Neural activity
Event related synchronization (red) and desynchronization in time-frequency
plots.
Attractors
Plot FSD representation of trajectories in 2 or 3 dimensions.
Attractors may change in time if voltage-dependent channels are present.
Depth of attractor basins
Variance around the center of a cluster grows with synaptic noise; for narrow
and deep attractors it will grow slowly, but for wide basins it will grow fast.
Jumping out of the attractor basin reduces the variance due to inhibition of
desynchronized neurons.
Model of reading
Emergent neural simulator:
Aisa, B., Mingus, B., and O'Reilly, R.
The emergent neural modeling
system. Neural Networks,
21, 1045-1212, 2008.
3-layer model of reading:
orthography, phonology, semantics,
or distribution of activity over 140
microfeatures of concepts.
Hidden layers in between.
Learning: mapping one of the 3 layers to the other two.
Fluctuations around final configuration = attractors representing concepts.
How to see properties of their basins, their relations?
Attractors for words
Dobosz K, Duch W, Fuzzy Symbolic Dynamics for Neurodynamical Systems.
Neural Networks (in print, 2009). Ex: 8 words, 4 abstract/4 concrete ones.
Attractors for words
Model for reading includes
phonological, orthographic and
semantic layers with hidden layers in
between.
Non-linear visualization of activity of
the semantic layer with 140 units.
Cost and rent have semantic
associations, attractors are close to
each other, but without
accommodation basins are small and
narrow.
Broadening phonological/written form
representations may help.
Will training of autistic children with increasing variance stimuli help?
Inhibition
Increasing
gi from 0.9 to 1.1
reduces the attractor
basins and
simplifies trajectories.
Connectivity
With small synaptic noise
(var=0.02) the network starts from
reaching an attractor and moves to
creates “chain of thoughts”.
Same situation but recurrent
connections within layers are
stronger, fewer but larger
attractors are reached,
more time is spent in each
attractor.
Some speculations
Attention shifts may be impaired due to the:
1.
Deep and narrow attractors that entrap dynamics – leak channels?
Explains overspecific memory in ASD, unusual attention to details, the
inability to generalize visual and other stimuli.
2.
Accommodation: voltage-dependent K+ channels (~40 types) do not
decrease depolarization in a normal way, attractors do not shrink.
This effect should also slow down attention shifts and reduce jumps to
unrelated thoughts or topics relatively to average person – neural fatigue will
temporarily switch them off preventing activation of attractors that code
significantly overlapping concepts.
What behavioral changes are expected? How to tests for them?
Consequences
Deep, localized attractors are formed; what are the consequences?
• Problems with disengagement of attention;
• precise memory for images, words, numbers, facts, movements;
• strong focus on single stimulus, absorption, easy sensory overstimulation;
• in motor cortex this leads to repetitive movements;
• generalization and associations are quite poor; integration of different
modalities that requires synchronization is impaired, connections are weak;
• echolalia, repeating words without understanding (no associations);
“has the name but not the meaning” … trapped in the sound;
nouns are acquired more readily than abstract words like verbs;
• play is schematic, fast changes are not noticed (stable states cannot arise);
• play with other children is avoided in favor of simple toys;
• faces are ignored (change to fast), and thus contact with caretakers is
difficult, gaze focused on simple stimuli;
• normal development – relations, theory of mind, empathy – is impaired.
Simple basic deficit => host of problems, severity and local expression,
many insights from simple but general mechanism.
Experimental evidence: behavior
Kawakubo Y, et al. Electrophysiological abnormalities of spatial attention in
adults with autism during the gap overlap task. Clinical Neurophysiology
118(7), 1464-1471, 2007.
• “These results demonstrate electrophysiological abnormalities of
disengagement during visuospatial attention in adults with autism which
cannot be attributed to their IQs.”
• “We suggest that adults with autism have deficits in attentional
disengagement and the physiological substrates underlying deficits in
autism and mental retardation are different.”
Landry R, Bryson SE, Impaired disengagement of attention in young children with
autism. Journal of Child Psychology and Psychiatry 45(6), 1115 - 1122, 2004
•
“Children with autism had marked difficulty in disengaging attention. Indeed,
on 20% of trials they remained fixated on the first of two competing stimuli
for the entire 8-second trial duration.”
Several newer studies: Mayada Elsabbagh.
Experimental evidence: molecular
What type of problems with neurons create these types of effects?
• Neural self-regulation mechanisms lead to fatigue or accommodation of
neurons through leak K+ channels opened by increasing Ca concentrations,
or longer acting GABA-B inhibitory synaptic channel.
• This leads to inhibition of neurons that require stronger activation to fire.
• Neurons accommodate or fatigue and become less and less active for the
same amount of excitatory input.
Dysregulated calcium signaling, mainly through voltage-gated calcium channels
(VGCC) is the central molecular event that leads to pathologies of autism.
http://www.autismcalciumchannelopathy.com/ & 45 Wikipedia entries.
Calcium homeostasis in critical stages of development may be perturbed by
genetic polymorphism related to immune function and inflammatory reactions
and environmental influences (perinatal hypoxia, infectious agents, toxins).
Genetic mutations => proteins building incorrect potassium channels
(CASPR2 gene) and sodium channels (SCN2A gene).
Genes & functions
I.N. Pessah, P.J. Lein, Evidence for Environmental Susceptibility in Autism
What We Need to Know About Gene x Environment Interactions , Ch. 13
Questions
There are many parameters characterizing biophysical properties of neurons
and their connections within different layers.
• How does depth/size of basins of attractors depend on these parameters?
How to measure and/or visualize attractors? Real effects or artefacts?
• How do attractors depend on the dynamics of neuron accommodation?
Noise? Inhibition strength, local excitations, long-distance synchronization?
• How will symptoms differ depending on specific brain areas? For example,
mu suppression may be due to deep attractors, brain compartmentalized …
• What are precise relations to ion channels and proteins that build them?
• Some parameters may be changed by pharmacological intervention, but also
learning procedures may have some influence on how these basins are
formed – for example, learning to read may depend on the variability of
fonts, handwriting may be much more difficult etc.
• Different ion channels => different temporal dynamics => associations?
• Can one draw useful suggestions how to compensate in such cases?
• Will it help in diagnostics and therapy?
Generative Psychiatry
• How genetic and molecular changes influence normal neurodynamics?
• What will be the effects of local changes in some parts of the brain only?
• What will be the effects of lack of synchronization between brain areas?
•
•
•
•
•
•
•
Consciousness levels:
The role of the brain stem (reticular formations) in regulation of cortex.
Transition from normal state to lowered states and to brain death.
Coma – no reactions to external stimuli.
Minimal consciousness state, with islands of activity left.
Vegetative state, only spontaneous movements and circadian rhythms.
Brain death process.
• Neuroanatomy and psychology of talent?
• Feedback projections (dt-MRI) to sensory cortices should be critical in
creation of vivid imagery.
Conference Series
• Body, perception and awareness. Motor and multimodal
perspectives, Toruń, 23-25.11.2009.
• http://www.kognitywistyka.net/~bpa/
Interdisciplinary conference following:
•
•
•
•
Enactivism: A new paradigm? From neurophenomenology and
social/evolutionary robotics to distributed cognition (2008).
Self, Intersubjectivity & Social Neuroscience: from Mind and Action to
Society (2007).
Embodied and Situated Cognition: from Phenomenology and
Neuroscience to Artificial Intelligence (2006).
Mirror neurons – from action to empathy, Toruń, From Action to
Empathy, 14-16.04.10
Thank
you
for
lending your
ears
...
Google: W. Duch => Papers, Talks