THE THORNY PATH TO AN ARTIFICIAL BRAIN

How to Build a Bridge between Neurophysiology and Network Modeling

Elena Saftenku

Department of General Physiology, A. A. Bogomoletz Institute of Physiology, 4 Bogomoletz St, Kyiv, Ukraine

Keywords: Neural Networks, Network Processing, Brain-based Device.

Abstract: Humanoid robots are created to imitate some of the tasks that humans undergo, but no current robot can

emulate the cognitive capabilities of even the simplest mammals. One approach to developing computing

platforms for cognitive robotics is to make use of experimental characterizations of the neurobiological

substrate for action and perception systems and simulate brain functions designing real-time spiking neural

networks. Biologically detailed network models are a powerful tool to understand how molecular and

cellular mechanisms determine high level network processing. Recent advances in experimental and

theoretical studies of the dynamic organization of neuronal populations suggest that our further success in

creation of higher intelligence robots will depend on the ability to incorporate such basic principles of brain

functioning as (i) stochastic dynamics and intrinsic nonlinearities in input-output transformation of neurons,

(ii) structural and functional plasticity, (iii) signaling through neuromodulator networks.

1 INTRODUCTION

The adult human brain contains about 86 billion

neurons (Azevedo, Carvalho, Grinberg, Farfel,

Ferretti et al., 2009). The main function of neurons

is to process and transmit information. This

transmission occurs via roughly 10,000 chemical

and, in few locations, electrical synapses. Both

synapses and neurons have complex stochastic

dynamic properties. The ability to learn and fulfill

fine movements is achieved after an integration and

representation in the brain of information from a

large number of sensorimotor and cognitive signals.

The human brain can rewire itself and change its

structure and function in response to an experience,

can pursue goals, think abstractly and creatively.

Serious consideration to the possibility of building

an electronic brain starts from the middle of the last

century. However, the traditional artificial neural

networks contain only very simplified (one-node)

models of biological neurons with reciprocal

interactions between all nodes and require a large

diversity of training for real-world operations. In

spite of immense calculation speeds, machines are

much less effective than biological systems in real-

world environments and cannot match the

capabilities of the human brain. At the same time, a

number of detailed network models representing

diverse types of neurons with their complex

morphology and distinct subsets of ion channels

were simulated (see Markram, 2006 for review).

These models are intended to fill the considerable

gap in our understanding between the processes on

the molecular level and on the level of network

function. It is sufficiently difficult to build network

models that incorporate realistic morphologies and

asynchronous, dynamical and self-organizing

changes in the synaptic connections and intrinsic

properties. However, it is evident that understanding

of the brain requires the development of realistic

detailed models based on experimental description

of all of their components that can be tested and

refined through new experiments. Recently a large

international project whose aim is to simulate in a

supercomputer the brains of mammals with a high

level of biological accuracy has been started

(Markram, 2006). This may help to create a new

generation of intelligent neuromorphic devices with

the ability to form neural representations of their

bodies and environment. Internal models will allow

them to plan and prepare for future eventualities. I

shall emphasize several aspects of neural

functioning that may be important for biologically

accurate brain simulations.

170

Saftenku E..

THE THORNY PATH TO AN ARTIFICIAL BRAIN - How to Build a Bridge between Neurophysiology and Network Modeling.

DOI: 10.5220/0003824901700176

In Proceedings of the 2nd International Conference on Pervasive Embedded Computing and Communication Systems (PECCS-2012), pages 170-176

ISBN: 978-989-8565-00-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 REQUIREMENTS FOR

BIOLOGICALLY ACCURATE

BRAIN SIMULATIONS

2.1 Intrinsic Nonlinearities in Input-

Output Transformation of Neurons

The most difficulty in accurately modeling signal

flow in neural circuits arises from the fact that the

electrical behavior of neurons is determined by a

large number of non-linear elements, such as

membrane ion channels with non-linear voltage-

dependence and synapses with highly non-linear

transmission. Moreover, most neurons have

extensive dendritic trees, but the distribution and

properties of dendritic ion channels still cannot be

characterized experimentally. Dendrites integrate

multiple synaptic inputs and translate them into

axonal action potentials (APs). The spiking of

neurons is usually driven by irregular synaptic inputs

and stochastic flickering of voltage-gated channels.

Physiological levels of channel noise can produce

qualitative changes in neural dynamics (Dorval and

White, 2005) and allow probabilistic synaptic

integration (Cannon, O’Donnell and Nolan, 2010).

However, synaptic transmission is often a major

contributor to the voltage noise. Traditionally, it was

thought that a neuron simply summates synaptic

inputs, which it receives. When physiologically

realistic patterns of synaptic inputs, as observed in

vivo, were included in in vitro experiments, it was

found that the slope of the relationship between

mean input and output firing rates (gain) is

fundamentally altered (Silver, 2010). For neurons

that operate in coincidence-detection mode under

sparse activation conditions (Olshausen and Field,

2004), synaptic noise results in the broadening of the

time window for synaptic integration. The temporal

fidelity of coincidence detection is influenced by

shunting inhibitory conductances, which can be

regulated by voltage-dependent membrane

conductances (Pavlov, Scimemi, Savtchenko,

Kullmann and Walker, 2011). In addition, neural

gain can be controlled by active dendritic

conductances under some conditions (Silver, 2010)

It is widely believed that activity-dependent

change in synaptic plasticity is a fundamental

mechanism for stably altering the function of neural

networks. The standard models of memory in

neuroscience are based on Hebb’s postulate (Hebb,

1949) that repetitive co-activation of neurons

strengthens the connection among them. These

models assume that only the synaptic strength

changes. However, short-term changes in the

dynamics of the synaptic transmission introduce the

frequency-dependent nonlinearity and modify the

sensitivity of a neuron to the temporal coincidence

of its inputs. These changes can be modulated by

long-term plasticity. In result, the strengthening of

synapses for different frequencies appears to be non-

uniform and depends on prior activity (Silver, 2010;

Markram, Gerstner and Sjöström, 2011). Changing

the gain of neurons alters their responsiveness to

input and, therefore, their functional connectivity in

the network (Haider and McCormick, 2009). This is

a mechanism by which functional neuronal

assemblies can be formed and broken. Therefore, in

order to understand the operating principles of

network dynamics in the brain, it is necessary to

include in the models realistic dynamics of synaptic

transmission and the spatial and temporal patterns of

synaptic activation observed in vivo.

Another important issue that has been paid

insufficient attention in the past is that neural coding

may be based on selective responses of neurons to

some subsets of input interpulse intervals

(Vartanian, Pirogov and Shabaev, 1986). Interpulse

intervals can carry sufficiently higher stimulus-

related information than either spike-timing

precision or mean firing rate (Imaizumi, Priebe,

Sharpee, Cheung and Schreiner, 2010). Specific

classes of voltage-gated currents support

subthreshold oscillations of the membrane potential

and the intrinsic frequency preferences of neurons

(Hutcheon and Yarom, 2000). Recently,

experimental recordings have demonstrated a critical

role of cell-specific subthreshold oscillations of

membrane potential for spatial firing of grid cells in

enthorhinal cortex (Giocomo, Zilli, Fransén and

Hasselmo, 2007).

An additional complexity stems from the

dependence of information transfer in neural circuits

not only on synaptic transmission, but also on the

diffusion of neurotransmitter molecules through the

extracellular and cerebrospinal fluid. For example,

hippocampal neurogliaform cells release the

inhibitory neurotransmitter γ-aminobuteric acid

(GABA) and do not require synapses to produce

inhibitory responses in the majority of nearby

neurons (Oláh, Füle, Komlósi, Varga, Báldi et al.,

2009). The ambient levels of GABA can increase or

decrease the neuron firing probability (Song,

Savtchenko, and Semyanov, 2010). Spillover of the

main excitatory neurotransmitter glutamate from the

synaptic cleft prolongs the decay of synaptic

currents, increasing the time window of synaptic

integration and may activate presynaptic

THE THORNY PATH TO AN ARTIFICIAL BRAIN - How to Build a Bridge between Neurophysiology and Network

Modeling

171

metabotropic glutamate receptors at inhibitory

neurons.

2.2 Multiple Forms of Experience-

Induced Plasticity in the Brain

One of the central goals of computational

neuroscience is to explain how learning and memory

is achieved in the brain. Learning algorithms in

existing models of neural nets use synaptic plasticity

rules derived from already existing synapses to

reorganize or to reconfigure the connectivity within

a group of neurons. However, experience-dependent

plasticity in adult neural circuits may involve

formation and elimination of synaptic contacts,

spines, and axonal boutons (Holtmaat and Svoboda,

2009). For example, in the neocortex, the induced

appearance and disappearance of multiple synaptic

contacts over a time scale of hours was directly

shown in experiments after glutamate application

(Le Bé and Markram, 2006). Spine and synapse

densities can increase after training in enriched

environment, after long-term sensory stimulation,

and after induction of long-term potentiation (LTP),

which is an artificial form of plasticity (Lambrecht

and LeDoux, 2004). Nevertheless, the determinants

of the location of new synaptic connections are not

clear. Moreover, exposing animals to complex

environment leads to rearrangement of

axonal/dendritic arbors (Galimberti, Gogolla, Alberi,

Santos, Muller et al., 2006). The actin cytoskeleton

plays a major role in structural changes of neurons.

It constantly rearranges in response to neuronal

activity; and this leads to formation of new axonal

varicosities and to changes in the head volume of

dendritic spines (Dillon and Goda, 2005). The

structural changes of the spine head result in

changes of the calcium dynamics that controls, in its

turn, the induction of synaptic plasticity. Besides,

actin may contribute to synaptic transmission as it is

involved in the trafficking of glutamate and GABA

receptors and in vesicle translocation (Cingolani and

Goda, 2008). A number of limitations preclude

realistic modeling. Simulation environments for

modeling individual neurons and neural circuits

provide tools for spatial models with biologically

realistic morphology and synaptic connections, but

this morphology must be fixed. Solving of

diffusion−reaction systems on domains with moving

boundaries is challenging. So the techniques of

multi-level simulations should be developed in

which the model switches from dynamic structural

changes to electrical activity.

On the other hand, it becomes more and more

evident that learning cannot be completely equated

with synaptic plasticity. For example, intrinsic

plasticity in neurons can be considered as a cellular

correlate of learning. Neuronal activity persistently

regulates plasma membrane ion channels, which

determine the ability of a neuron to generate an AP

in response to a given input signal. Persistent

changes in the intrinsic excitability of neurons

elicited by modifications in the properties and/or

number of ion channels can be produced by training

in behaving animals or by activation of cellular

preparations by definite artificial patterns. These

changes may influence modifications in the synaptic

strength in a defined time window following the

training, function as a part of the engram itself,

promote the consolidation of memory, and

contribute to saving during reacquisition or to cross-

modal acquisition (Zhang and Linden, 2003). The

activity-dependent modulation of synaptic plasticity,

the so-called metaplasticity (Abraham and Bear,

1996), can be caused by an alteration of the

threshold for axosomatic spike generation or by a

change of the properties of voltage-gated channels in

a local domain of the dendritic tree. The latter is

especially important since most neuronal types have

a remarkable dendritic arbor onto which the majority

of synaptic connections are made. Moreover, the

properties and localization of dendritic voltage-gated

channels can be altered by synaptic activity or

neuromodulators and result in the qualitative change

of the neuronal firing patterns (Remy, Beck and

Yaari, 2010). Thus, both neurons and synapses are

history-dependent, and learning occurs at multiple

levels and time scales.

In addition, the brain responds to experience by

adding new neurons, glial cells and capillaries

(Grossman, Churchill, Bates, Kleim and Greenough,

2002). Induction of LTP at excitatory synapses

depends on signalling molecules released by

astrocytes (Henneberger, Papouin, Oliet and

Rusakov, 2010). Therefore, the simulations should

include the glial networks to capture neuron-glia

interactions.

2.3 Neuromodulatory Control of

Synaptic Transmission and

Neuronal Excitability

The most difficult problem of neuroscience is to

understand how system-level brain functions may

arise from low level molecular and cellular

mechanisms. The brain is connected with the body

and the body and brain interact with the external

environment. The behavior of an animal or human at

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

172

each given moment is hierarchically organized and

directed towards the satisfaction of some need,

which predominates over all other needs. Needs can

be biological (e.g., for meals), social (e.g., for

communication), and cognitive (e.g. for novelty).

While a need induces active behavior, the acquired

connections between a need and an object, which

can it satisfy, makes the behavior and learning

reward-mediated. The satisfaction of needs induces

positive emotional states.

Neuromodulators, the substances that alter the

function of other neurons at a slower time scale than

neurotransmitters and diffuse through large areas,

have a vital role in regulation of cognitive processes

and behavior. They change intrinsic excitability of

neurons, presynaptic release of neurotransmitters,

and the conditions for induction of long-term

synaptic plasticity (Schweighofer, Doya and Kuroda,

2004; Hasselmo and Sarter, 2011; Pawlak, Wickens,

Kirkwood and Kerr, 2010). Neuromodulators are

thought to be essentially required for induction of

spike-dependent plasticity at specific synapses in

vivo where there is a huge amount of constantly

ongoing presynaptic and backpropagating spiking

activity (Pawlak et al., 2010). Some

neuromodulators, e.g. such as dopamine,

noradrenaline, acetylcholine, serotonin, opioid

peptides, are involved in behaviorally based learning

and reward and play a particular important role in

emotional responses. Thus, dopamine provides

reward prediction errors and its release may be

activated by reward-predicting stimuli (Schultz,

2010). However, it is still unclear how the specific

neuronal activity around the reward event links to

the behavioral outcome (Pawlak et al., 2010). A

possible mechanism determining the crucial role of

emotional reinforcement in formation of the

functional connectivity between neurons could be

gain modulation to sensory stimuli due to tonic

changes in membrane potential. These changes may

be evoked by the release of neuromodulators

(Vartanian et al., 1986). To make matters more

complicated, we know that certain brain structures

process more specific reward information and

predictions of future outcomes, but our general

knowledge about how the reward systems are

organized is very incomplete.

The majority of monoaminergic neurons do not

make synaptic contacts and release neuromodulators

for long-distance diffusion. Some other locally

acting systems, such as endocannabinoids,

metabotropic glutamate receptors, brain-derived

neurotrophic factor (BDNF), and retrograde

messengers also play an important role in synaptic

plasticity. BDNF modulates synaptic transmission

and membrane excitability and is thought to be

necessary and sufficient for long-term memory

retention in the hippocampus (Cunha, Brambilla and

Thomas, 2010). Taking into consideration that a

variety of neuromodulatory agents are released at

different concentrations in behaving animals and

may interact, realistic modeling can be a very

challenging task.

One of the most intriguing issues is the role of

activity-dependent plasticity in forming assemblies

of neurons with pre-specified genetically determined

connectivity. Recently it was shown that small

clusters of pyramidal neurons in the neocortex

containing about 50 neurons make predictable

connections with predictable synaptic weights

independently of individual experiences (Perin,

Berger and Markram, 2011). Connection probability

between any two neurons increases linearly with the

number of their common neighbors. This

synaptically organizing principle is genetically

prescribed and applies across different animals. It

was suggested that acquired memory relies on

combining these microcircuits, which are

fundamental building blocks of perception (Perin et

al., 2011). The theory of neuronal group selection

(Edelman, 1993) maintains that the brain gives

repertoires of variant neuronal groups. The groups

that emerged during embryonic development are

selected to match the novelty and diversity of

experience under control of inborn value systems

producing neuromodulators. Experience could serve

to combine these groups in a hierarchical manner.

3 FUTURE DIRECTIONS

Henry Markram, the founder of Brain Mind Institute

in Switzerland, has claimed that with the right

resources and strategy it would be possible to

simulate the complete human brain at the cellular

level within 12 years. The Human Brain Project

proposes to integrate everything that we know about

brain into computer models and use these models to

simulate the actual working of the brain on a

supercomputer (http://www.humanbrainproject.eu).

It looks likely that the main constraint for this

project may be not an insufficient power of

supercomputer, which grows very quickly, but

insufficient experimental data necessary for model

development.

As it was mentioned above, one of the most

crucial issues in neuroscience is to establish the

functional role and mechanisms of neuromodulatory

THE THORNY PATH TO AN ARTIFICIAL BRAIN - How to Build a Bridge between Neurophysiology and Network

Modeling

173

control in cortical structures. For this, the

combination of in vitro and in vivo techniques is

required to bridge between cellular effects and

behavioral functions. Most studies of neural

dynamics and plasticity have been carried out in

vitro. Systematic recordings in vivo are technically

very demanding, they can be carried out only from

the largest neuronal elements. Instead, optical and

electrophysiological recordings are performed using

reduced preparations, particularly acute brain slices,

where controlled analysis of neuronal activity and

cellular properties can be made. In in vitro slice

preparations many synaptic connections are cut and

natural neuromodulation does not occur at all. In

result, the majority of experiments are conducted

under conditions where high levels of stochastic

voltage noise, which are observed in vivo, are

significantly decreased and the degree of

neuromodulation is negligible. However, both

synaptic noise and changes in neural excitability

evoked by neuromodulators alter the way the neuron

transforms its synaptic input into output firing rate

(Silver, 2010). Future optogenetic work might

delineate the contributions of the different

components of the cellular responses elicited by

neuromodulators to behavioral learning and identify

the particular forms of learning sensitive to these

substances.

The appropriate level of physical detail required

to understand how the behavioral function emerges

from the observed effects at the molecular and

cellular levels is unclear. Some phenomenological

descriptions and simplifications are inevitable

because of the limitations of realistic modeling, but

it is clear that such a model may not replicate

faithfully the neuron’s dynamics under different

conditions. If the prediction of the model does match

experimental data, it does not guarantee the validity

of the model, but should suggest new predictions

that can be verified experimentally or other

experiments that can test its validity under different

conditions. This approach drew on the rich history of

biophysical research and may be used for models at

different levels of complexity.

Presently, our group is developing a large-scale

computational model of the cerebellum based on

some recent experimental data to show that learning

in this brain structure can be regulated rather by

neuromodulators and neuropeptides than by a

climbing fiber error-driven teaching signal.

Many important details still remain to be

specified. The involvement of long-term synaptic

plasticity in learning and memory remains to be

conclusively demonstrated. We still do know neither

what cellular processes are necessary for the

maintenance and retrieval of long-term memory nor

what processes are central to the persistence of

memory after recall. Recent finding of innate neural

cortical assemblies (Perin et al., 2011) suggests that

in order to construct neural microcircuits in the

neocortex with realistic properties, it is necessary to

create at first these assemblies using genetically

determined connectivity principles and then to apply

a learning rule to associate them. But we know a

little about the topography of neurons in other brain

areas. Interestingly, there is convincing evidence

that the autoassociator theory of memory (Hebb,

1949) is incorrect for the hippocampus where the

mutual synaptic interconnections are set up in early

development (Colgin, Leutgeb, Jezek, Leutgeb,

Moser et al., 2010).

The greatest challenges, however, appear when

higher brain functions, such as cognition or

consciousness are attempted to be reproduced. The

subjective experience of each living being is unique

and arises from the trinity of the brain, body, and

environment. The brain works as a whole system,

and only one percept at any time is possible. Each

percept involves many brain areas simultaneously in

order to update episodic memory, spatial maps,

value systems, prefrontal planning, and motor

preparation (Edelman, Gally and Baars, 2011). Our

knowledge about the function and connections

between different brain structures are still

incomplete. For example, neuroscience inquires of

such a basic human ability as creativity show a

muddled picture. On the other hand, the construction

of brain-based devices, which should incorporate the

main principles of brain functioning and value

system, can help us to explore how numerous

biological mechanisms may interact to create new

system properties.

4 CONCLUSIONS

Computer modeling is becoming a valuable tool for

understanding how high brain functions arise from

molecular and cellular mechanisms. The living brain

is much more complex of any brain-based device,

which it is possible to imagine. We only have begun

to understand some basic operating characteristics of

neural networks. Control of neural dynamics and

connectivity by synaptic noise, a combination of

biochemical networks of neuromodulators with

neural networks to perform computations, the

existence of multiple forms of plasticity are

fundamental principles of their functioning. These

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

174

principles should be included in future models. Our

basic knowledge about neural coding and brain

functions on the systems level are still very

insufficient. Hopefully, computational science and

neuroscience will develop with a close

interdependence, such that model predictions will

inspire new experiments with discrepancies between

theory and experiment serving as the impetus for

model refinement.

REFERENCES

Abraham, W. C., Bear, M. F. (1996). Metaplasticity: the

plasticity of synaptic plasticity. Trends Neurosci, 19,

126-130.

Azevedo, F. A., Carvalho, L. R., Grinberg. L. T., Farfel, J.

M., Ferretti, R. E., Leite, R. E. (...), Herculano-Houzel,

S. (2009). Equal numbers of neuronal and nonneuronal

cells make the human brain an isometrically scaled-up

primate brain. J Comp Neurol, 513, 532-541.

Cannon, R. C., O’Donnell, C. O., Nolan, M. F. (2010).

Stochastic ion channel gating in dendritic neurons:

morphology dependence and probabilistic synaptic

activation of dendritic spikes. PLoS Comput Biol, 6,

e1000886.

Cingolani, L. A., Goda, Y. (2008) Actin in action: the

interplay between the actin cytoskeleton and synaptic

efficacy. Nat Rev Neurosci, 9, 344-356.

Colgin, L. L., Leutgeb, S., Jezek, K., Leutgeb, J. K.,

Moser, E. I., McNaughton, B. L., Moser, M. B.

(2010). Attractor-map versus autoassociation based

attractor dynamics in the hippocampal network. J

Neurophysiol, 104, 35-50.

Cunha, C., Brambilla, R., Thomas K. L. (2010). A simple

role for BDNF in learning and memory? Front Mol

Neurosci, 3, 1.

Dillon, C., Goda, Y. (2005). The actin cytoskeleton:

integrating form and function at the synapse. Annu Rev

Neurosci, 28, 25-55.

Dorval, A. D., White, J. A. (2005). Channel noise is

essential for perithreshold oscillations in entorhinal

stellate neurons. J Neurosci, 25, 10025-10028.

Edelman, G. M. (1993). Neural Darwinism: selection and

reentrant signaling in higher brain function. Neuron,

10, 115-125.

Edelman, G. M., Gally, J. A., Baars, B. J. (2011). Biology

of consciousness. Front Psychol, 2, 4.

Galimberti, I., Gogolla, N., Alberi, S., Santos, A. F.,

Muller, D., Caroni, P. (2006). Long-term

rearrangements of hippocampal mossy fiber terminal

connectivity in the adult regulated by experience.

Neuron, 50, 749-763.

Giocomo, L. M., Zilli, E. A., Fransén, E., Hasselmo, M. E.

(2007). Temporal frequency of subthreshold

oscillations scales with entorhinal grid cell field

spacing. Science, 315, 1719-1722.

Grossman, A. W., Churchill, J. D., Bates K.E., Kleim, J.

A., Greenough, W. T. (2002). A brain adaptation view

of plasticity: is synaptic plasticity an overly limited

concept? Prog Brain Res, 138, 91-108.

Haider, B., McCormick, D. A. (2009). Rapid neocortical

dynamics: cellular and network mechanisms. Neuron,

62, 171-189.

Hasselmo, M. E., Sarter, M. (2011). Modes and models of

forebrain cholinergic neuromodulation of cognition.

Neuropsychopharmacology, 36, 52-73.

Hebb, D. O. (1949). The Organization of Behavior. New

York: Wiley.

Henneberger, C., Papouin, T., Oliet, S. H., Rusakov, D. A.

(2010). Long-term potentiation depends on release of

D-serine from astrocytes. Nature, 463, 232-236.

Holtmaat, A., Svoboda, K. (2009). Experience-dependent

structural synaptic plasticity in the mammalian brain.

Nat Rev Neurosci, 10, 647-658.

Hutcheon, B., Yarom, Y. (2000). Resonance, oscillation

and the intrinsic frequency preferences of neurons.

Trends Neurosci, 23, 216-222.

Imaizumi, K., Priebe, N. J., Sharpee, T. O., Cheung, S.

W., Schreiner, C.E. (2010). Encoding of temporal

information by timing, rate and place in cat auditory

cortex. PLoS One, 5, e11531.

Lambrecht, R., LeDoux, J. (2004). Structural plasticity

and memory. Nat Rev Neurosci, 5, 45-54.

Le Bé, J. V., Markram, H. (2006). Spontaneous and

evoked synaptic rewiring in the neonatal neocortex.

Proc Natl Acad Sci USA, 103, 13214-13219.

Markram, H. (2006). The Blue Brain project. Nat Rev

Neurosci, 7, 153-160.

Markram, H., Gerstner, W., Sjöström, P. J. (2011) A

history of spike-timing-dependent plasticity. Front Syn

Neurosci, 3, 4.

Oláh, S., Füle, M., Komlósi, G., Varga, C., Báldi, R.,

Barzó, P., Tamás, G. (2009). Regulation of cortical

microcircuits by unitary GABA-mediated volume

transmission. Nature, 461, 1278-1281.

Olshausen, B. A., Field, D. J. (2004). Sparse coding of

sensory inputs. Curr Opin Neurobiol, 14, 481-487

Pavlov, I., Scimemi, A., Savtchenko, L., Kullmann, D. M.,

Walker, M. C. (2011). I

-mediated depolarization

enhances the temporal precision of neuronal

integration. Nat Commun, 2, 199.

Pawlak, V., Wickens, J. R., Kirkwood, A., Kerr, J. N.

(2010) Timing is not everything: neuromodulation

open the STDP gate. Front Syn Neurosci, 2, 146.

Perin, R., Berger, T. K., Markram, H. (2011). A synaptic

organizing principle for cortical neuronal groups. Proc

Natl Acad Sci USA, 108, 5419-5424.

Remy, S., Beck, H., Yaari, Y. (2010). Plasticity of

voltage-gated ion channels in pyramidal cell dendrites.

Curr Opin Neurobiol, 20, 503-509.

Schultz, W. (2010). Multiple functions of dopamine

neurons. F1000 Biol Rep, 2, 2.

Schweighofer, N., Doya, K., Kuroda, S. (2004). Cerebellar

aminergic neuromodulation: towards a functional

understanding. Rrain Res Rev, 44,103-116.

Silver, R. A. (2010). Neuronal arithmetic. Nat Rev

Neurosci, 11, 474-489.

Song, I., Savtchenko L., Semyanov, A (2011). Tonic

THE THORNY PATH TO AN ARTIFICIAL BRAIN - How to Build a Bridge between Neurophysiology and Network

Modeling

175

excitation or inhibition is set by GABA

conductance

in hippocampal interneurons. Nat Commun, 2, 376.

Vartanian, G. A., Pirogov, A. A., Shabaev, V. V. (1986).

Hypothesis of the information switching of

conditioned reflex activity. Zh Vyssh Nerv Deiat Im I

P Pavlova, 36, 1168-1179.

Zhang, W., Linden, D. J. (2003). The other side of

engram: experience–driven changes in neuronal

intrinsic excitability. Nat Rev Neurosci, 4, 885-900.

PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems

176