abstracts by niusheng11



      NAME                                 ABSTRACT TITLE
Takaoki Kasahara    Molecular Basis of Bipolar Disorder
Shangkai Gao        EEG Pattern Analysis and its Application in Brain Computer

Nihar Ranjan Jana   Co-chaperone CHIP associates with expanded polyglutamine
                    protein and promotes their degradation by proteasomes

Soo-Young Lee       A Roadmap to the Artificial Brain and „OfficeMate‟
Min W. Jung         Learning and memory in the prefrontal cortex
Shi-Yong Huang      Pre- and post-synaptic factors both contribute to the modification
                    of synapses between retinal neurons

Manho Kim           Huntington‟s disease: molecular mechanisms of
                    neurodegeneration and treatment for brain repair

Toru Aonishi        Effect of Backpropagating Action Potential on Neural Interaction
Latika Singh        Variability in Spectro-Temporal Features: a Developmental Study
                    of Speech in Children

Qinye Tong          How to Understand Neural Coding?

Key-Sun Choi        A Computational Causality for Explanatory Language
Masumi Ishikawa     A new approach to localization and navigation of mobile robots -
                    Effective Bayesian estimation and reinforcement learning.

Kyungjin Kim        Neurobiology of Stress
Prasun Roy          Can Stochastic Resonance Imaging be a substitute for high-priced
                    Gadolinium scan in India?

Bao-Ming LI         Regulation of Prefrontal Cortical Functions by Alpha-2-
                    adrenoceptors: Its possible relevance to Attention Deficit
                    Hyperactivity Disorder
Teiichi Furuichi    Deciphering the genetic basis of mouse cerebellar development
Prabodh Swain       MAPK regulates phosphorylation of Neural Retina Leucine
                    Zipper: A Key Regulator of Rod Photoreceptor Differentiation
                    and Function
Tao Zhang             Examine synchrony of the relationship between blood pressure
                      (BP) and renal sympathetic nerve activity (RSNA) in response to
                      haemorrhage in Wistar rats.Examine synchrony of the relationship
                      between blood pressure (BP) and renal sympathetic nerve activity
                      (RSNA) in response to haemorrhage in Wistar rats.
Xin Tian              From Neural Firing to Hypersynchronous Discharge in the Cortex
                      of Epileptogenic Focus

Hiroyuki Kamiguchi    The Role of Ankyrins in Neurite Growth and Polarization

Kiyotoshi Matsuoka    Blind Separation of Sound Sources in Real-World Situations

Supriya Ray           Neural Control of Saccade Sequences
Yisheng Zhu           Monitoring Ischemic Brain Injury Using Nonlinear Methods
Jong-Hye Han          Neuroanatomical Analysis for Onoamtopoeia and Phainomime W
                      ords:    fMRI Study


     NAME                                    POSTER
Ling Yin             Neuroinformatics and Data Sharing
Keiji Kamei          Improvement of reinforcement learning of a mobile robot
                     using sensors and a genetic algorithm
Paulito Palmes       Robustness, Evolvability, and Optimality of Evolutionary
                     Neural Networks
Choong-Myung         Korean Sentence Processing Mechanisms reflected on
Kim                  ERP patterns
Kyoungho Suk         Signal transduction of auto-regulatory microglial apoptosis
Kyung-Joong Kim      Informal Inference based on the Integration of Multiple
                     Neural Networks
Michelle Jeungeun    Unsupervised Extraction of Video Features for Lipreading
Su-Yong Eun          Glutamate receptor-mediated signaling and the functional
                     implication in microglial cells
Woong Sun            Programmed cell death of adult generated hippocampal
                     neurons is mediated by the pro-apoptotic gene Bax
Reddy P.             Biotransformation of drugs mediated by brain-specific
Kommaddi             splice variants of the drug-metabolizing enzyme,
                     Cytochrome P450
Molecular Basis of Bipolar Disorder

                        Takaoki Kasahara, Tadafumi Kato
              Laboratory for Molecular Dynamics of Mental Disorders
                      RIKEN Brain Science Institute, Japan

       Bipolar disorder, previously known as manic depressive illness, affects
approximately 0.8% of the population and causes severe psychosocial impairment.
High concordance rate of bipolar disorder in monozygotic twins (approximately 70%)
suggests the role of genetic factors in this disorder.

        Altered energy metabolism in the brains of patients with bipolar disorder has
been reported using phosphorus magnetic resonance spectroscopy (31P-MRS). Some
of the 31P-MRS findings in bipolar disorder, such as reduced phosphocreatine levels,
were similar to those reported in patients with a mitochondrial disorder, i.e.
mitochondrial myopathy, chronic progressive external ophthalmoplegia (CPEO).
Autosomal dominantly inherited CPEO is caused by deletion of mitochondrial DNA
(mtDNA) due to the mutations in the nuclear genome-encoding genes responsible for
maintenance of mtDNA; mtDNA polymerase (POLG), adenine nucleotide
translocator (ANT1), and mtDNA helicase (TWINKLE). All three types of diseases
are accompanied by depression or bipolar disorder. Multiple deletions of mtDNA in
the brain were also reported in Wolfram disease, which is another inherited disease
and frequently accompanies bipolar disorder. These findings altogether suggested
that bipolar disorder is related to mitochondrial dysfunction caused by accumulation
of mtDNA deletions.

        To clarify the role of mitochondrial dysfunction in bipolar disorder, we are
currently studying pathology of bipolar disorder using molecular genetic and cellular
physiological approaches. We identified several mtDNA mutations/polymorphisms
significantly associated with bipolar disorder. Among them, 3644TC mutation
(odds ratio = 10.4) altered a conserved amino acid residue in one subunit of the
respiratory chain machinery, and cells containing the mtDNA mutation had reduced
mitochondrial membrane potential. We also generated genetically engineered mice
with neuron-specific accumulation of mtDNA mutations and deletions. Some
preliminary results demonstrated that the mutant mice showed bipolar disorder-like
symptoms. Therefore the mice will help us to understand the pathophysiology of
bipolar disorder as well as to develop mood-stabilizing drugs.
  EEG Pattern Analysis and its Application in Brain Computer Interface

    Shangkai Gao, Xiaorong Gao, Zhiguang Zhang, Bo Hong, Fusheng Yang
Department of Biomedical Engineering, Tsinghua University, Beijing 100084, China

        A brain-computer interface (BCI) is a communication channel directly
connecting the brain to a computer or other external devices. BCI will be very useful
for the people with severe motion disabilities.

        A variety of brain signals, such as scalp EEG, cortex EEG or direct neuron
activity, could serve as the original messages in a BCI system. Among them, scalp
EEG-based BCI systems are probably the most acceptable systems for practical use
because of its non-invasive property.

        However, the EEG signals recorded from the scalp are very week with low
signal-to-noise ratio and low spatial resolution due to the low conductivity of the skull
and the ambient noises. It seriously restricts the development of practical BCI
systems. It is commonly agreed now that the electric activities in the brain are
functionally located at different regions and they are synchronized in a cooperative
way. To extract task-related information buried in multi-channel scalp EEG, we
propose several spatio-temporal pattern analysis methods for scalp EEG analysis. In
this paper, an integrated framework of scalp EEG signal processing in BCI system is
presented. The algorithms of signal pre-processing, feature extraction and
classification are described in detail. Many of them have successively applied in the
BCI system developments.
Co-chaperone CHIP associates with expanded polyglutamine protein and
             promotes their degradation by proteasomes

  Nihar Ranjan Jana1, Priyanka Dikshit1, Anand Goswami1 and Nobuyuki Nukina2
  Cellular and Molecular Neuroscience Laboratory, National Brain Research Centre,
                    Manesar, Gurgaon - 122 050,Haryana, India
       Laboratory for Structural Neuropathology, RIKEN Brain Science Institute,
                 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan.

         A major hallmark of the polyglutamine diseases is the formation of neuronal
intranuclear inclusions (NIIs) of the disease proteins that are ubiquitinated and often
associated with various chaperones and proteasome components. But, how the
polyglutamine proteins are ubiquitinated and degraded by the proteasomes are not
known. Here, we demonstrate that CHIP (C-terminus of hsp70 interacting protein) co-
immunoprecipitates with the polyglutamine-expanded huntingtin or ataxin-3 and
associates with their aggregates. Transient over expression of CHIP increases the
ubiquitination and the rate of degradation of polyglutamine-expanded huntingtin or
ataxin-3. Finally, we show that over expression of CHIP suppresses the aggregation
and cell death mediated by expanded polyglutamine proteins and the suppressive
effect is more prominent when CHIP is over expressed along with Hsc70.
               A Roadmap to the Artificial Brain and „OfficeMate‟

                                 Soo-Young Lee
Brain Science Research Center, Department of BioSystems, and Department of EECS
            Korea Advanced Institute of Science and Technology, Korea

       The Korean Brain Neuroinformatics Research Program got into the 3rd phase
from June 2004 for 4 years, which is regarded as the final phase of Korean brain
national research program started from November 1998 for 10 years sponsored by
Ministry of Science and Technology. In the 3rd phase, in addition to the continuing
development of engineering models on functional artificial systems for vision,
auditory, inference, and behavior, we would like to develop a “Artificial Brain” to
combine all the 4 functions and an integrated demonstration system, i.e., “Artificial
Secretary” alias “OfficeMate” with exceptional human-like information processing

        We will develop an integrated hardware and software platform for the brain-
like intelligent systems, which combine all the technologies developed for the brain
functions in the second phase. With 2 microphones, 2 cameras (or retina chips), and
one speaker, the Artificial Brain will be able to see, hear, speak, and think. The sound
localization, speech enhancement, visual attention, face recognition, lip reading, and
emotion recognition and representation will be included. The Artificial Brain will
have proactive self-learning capability to develop his/her intelligence by estimating
his/her current status, asking good questions to the right persons, and incorporating
the answers into his/her knowledge. The ability of user modeling will also be included
for practical user interfaces.

        With this platform, we plan to develop a testbed application, i.e.,
“OfficeMate.” The OfficeMate may be regarded as an Artificial Brain specifically
trained for the job of secretarial and administrative works. He/she will be able to
receive phone calls, adjust schedules, and prepare draft documents for me.

        The scientific and technical challenges are enormous. The understanding on
brain information processing mechanism is very very limited, and we will need
exercise our imagination and educated guesses to fill up the missing holes. During the
last 6 years we have developed information-theoretic algorithms for the biological
information processing in the human auditory pathway. We will further extend this
approach to the audio-visual processing, knowledge representation and processing,
and mental development through proactive learning.

Intelligence to Machines! Freedom to Mankind!
            Learning and memory in the prefrontal cortex

                             Min W. Jung
         Neuroscience Laboratory, Institute for Medical Sciences,
        Ajou University School of Medicine, Suwon 443-721, Korea

       Modifying behavioral strategies in accordance with changes in
environment is extremely important for survival. The prefrontal cortex
(PFC) is likely to play a crucial role in this adaptive process considering
that one important function of the PFC is the planning of future
behaviors. In order to investigate neural mechanisms by which the PFC
adaptively modifies its activities based on past experience, we investigated
learning-induced changes in neural activity and synaptic plasticity in rat
PFC. Single neuron recording studies in behaving animals revealed that
PFC neural activities change rapidly in parallel with behavior learning.
Moreover, correlations among neurons were altered in the process of
learning, and long-term potentiation was induced by high-frequency
stimulation in sensory cortical projections to the PFC. These results
suggest that synaptic weights are modified within the PFC in the process
of new task learning so that neural activity changes dynamically as an
animal learns a new behavioral strategy. Same neurons exhibited
different activity patterns but correlations among neurons were similar
across two different behavioral tasks, suggesting that multiple behavioral
strategies are represented in an overlapping, distributed manner in the
PFC neural network. These studies stress the importance of learning and
memory as an essential component of the PFC functions.
         Role of GAP-43 in Differentiation of Cerebellar Granule Cells

                         R Mishra, K F Meiri# and S Mani
    National Brain Research Centre, Manesar,Gurgaon-122050, Haryana, India
                  Tufts University School Of Medicine, Boston, USA

        GAP-43 is a nervous system specific protein that plays an important role in
regulating growth cone responses to extracellular signals. GAP-43 is expressed as
early as E9.5 in mouse, which implicates its possible role in neurogenesis. All GAP-
43 (-/-) mice display abnormal foliation of specific lobules at the cerebellar vermis, a
defect that is apparent from the day of birth. Further, GAP-43 knockout mice have
smaller cerebella and reduced size of external granule cell layer. We investigated
whether the reduction in EGL was due to reduced cell proliferation. Using whole
cerebellar cultures from P4 wild type and knockout mice we show that there is a two-
fold increase in BrdU labeled cells in the knock out cultures as compared to the wild
type cerebellar cultures. We also show that proliferation response to bFGF when
added to the defined culture medium is greater in wild type cerebellar cultures than
for the knockout cultures. Similarly the proliferation response to Shh, an important
factor that regulates granule cell proliferation in the developing cerebellum, was also
reduced in the knockout cultures as compared to wild type cultures. We are currently
testing the hypothesis that the increase in BrdU labeled cells seen in the knockout is
due to a pertubation in cell cycle length. Taken together with the decreased response
of knock out cerebellar cells to extracellular signaling molecules, this in turn could
lead to abnormal EGL formation and foliation. We are also investigating whether
increased apoptosis in the absence of GAP-43 could contribute to a reduced EGL in
the knockouts.

Supported by:         1R03TW006050 (Fogarty International Award) KFM & SM
   Pre- and post-synaptic factors both contribute to the modification of
                    synapses between retinal neurons

                          Shi-Yong Huang1 & Pei-Ji Liang2,CA
      1. Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences
                    320 Yue-Yang Road, Shanghai 200031, China
       2. Department of Biomedical Engineering, Shanghai Jiao Tong University
                   1954 Hua-Shan Road, Shanghai 200030, China

        There are three types of cooperative phenomena in synapses between cones
and horizontal cell found in our laboratory. 1) Repetitive red flashed increase the R/G
horizontal cell‟s response amplitude to red light; 2) Repetitive red flashes
progressively strengthened the synaptic connection between red-cone and luminosity-
type horizontal cell (LHC); 3) A dim red background light greatly enhanced the
response amplitude of LHC to green test flash, and vice versa. Present study
suggested that the mechanisms involved in these phenomena were different between
each other. Pharmacological evidence indicated that the activation of D2 receptor
might be responsible for the red flickering induced responsiveness enhancement in the
R/G HC, and that the potentiation of red response in LHC is attributed to the change
of the postsynaptic intracellular calcium concentration, and that the mutual color
enhancement in LHC is related to glutamate transporter and the activation of mGluRs
located at the presynaptic terminals of cone. The synapses between photoreceptors
and LHC also have a competitive side: when synapses between red cone and LHC
grown stronger and prosper after repetitive red flashes, other synapses between green
cone and LHC were instead weakened, and vice versa. However, there is still no
evidence to clarity the mechanism(s) involved in the competitive suppression in LHC.
Taken together, red and green cone afferents interact on the horizontal cell, and
different cellular mechanisms are involved in the competition and cooperation.

Acknowledgements: This research was supported by grant from the National
Foundation of natural Science of China (No. 30170263).
  Huntington’s disease: molecular mechanisms of neurodegeneration and
                        treatment for brain repair

                                   Manho Kim
          Department of Neurology Seoul National University, Seoul, Korea

       Huntington's disease (HD) is a fatal, genetically based neurodegenerative
brain disorder in which a loss of neostriatal neurons is a main characteristic. The
CAG trinucleotide repeat mutation encoding an expanded polyglutamine tract induces
progressive deficits in intra- and inter-cellular signaling, and subsequent disease

        Altered protein interaction has been first proposed as the main mechanism of
neurodegeneration. Thereafter, intranuclear or intracellular aggregates, proteolytic
cleavage of huntingtin (cf. caspase, calpain), altered transcription or other signalling
deficits were reported. Recently, stem cell transplantation is of benefit to protect
neurons against neurodegeneration as well as recover the functional deficit in the
experimental HD model.

This presentation focuses on the current knowledge of molecular mechanisms of
neurogeneration in HD and the use of this information to identify potential therapeutic
     Effect of Backpropagating Action Potential on Neural Interaction

       Toru Aonishi 1, Hiroshi Miyakawa 2, Masashi Inoue 2, Masato Okada 3
          Tokyo Institute of Technology, RIKEN Brain Science Institute, Japan
                     Tokyo University of Pharmacy and Life Science
              RIKEN Brian Science Institute, University of Tokyo, PRESTO

        It is a serious problem that the information could not be shared with
physiologists and theorists in the brain science. Theorists and physiologists do not
have the common language to express the nerve system, because they have different
backgrounds. In our research group, by sharing mathematical neuron models through
a common platform refereed to as NEURON simulator, we try to collaborate with
those who have different backgrounds. By such new style of collaborations, we
elucidate roles of the dendritic backpropagating action potential in the neural

        The membrane can be modeled by the parallel circuit that consists of inward
currents and outward currents with nonlinear I-V relations. If the balance between the
inward current and the outward current is satisfied, there is a singular point where the
effective membrane impedance diverges to infinity. This kind of singularity is
universal in nerve membranes. By using NEURON simulator, we show a dramatic
change of EPSPs by regulations of outward currents in backpropagating states. This
phenomenon reflects the singularity of the membrane impedance. From a picture of
such extreme phenomenon, we can elucidate roles of the dendritic backpropagation in
the neural interaction. Next, we observe phase response curves of a neuron in
oscillatory states. This experiment includes previous experiments for EPSP
modulations in backpropagating states, and thus the effect of such extreme
phenomenon on network dynamics can be elucidated.
   Variability in Spectro-Temporal Features: a Developmental Study of
                            Speech in Children

                       Latika Singh and Nandini C. Singh
    National Brain Research Centre, Manesar, Gurgaon-122050, Haryana, India

        We carried out a study to examine the variability in the acoustic properties of
speech for normally developing children. We measured and compared the joint
spectral and temporal features by estimating modulation spectra. These modulation
spectra were obtained by calculating the two-dimensional Fourier transform of the
autocorrelation matrix of the speech signal in its spectrographic representation.
Various parameters were determined to quantitatively study the changes in these
features and results revealed significant differences that might be related to learning
that occurs during development. The results show increase in separability as a major
trend associated with speech development in normal children. Between ages 3 and 5,
separability increases rapidly moving towards adult speech, which is quite separable.
Reduction in modulation depth with age is another important parametric change. It is
in accordance with previous studies showing decrease in temporal and spectral
parameters during development. The examination of such characteristics in children
provides us with information on how these parameters of speech are acquired and
mastered as a function of development and could lead to newer insights to
corresponding anatomical and neuromuscular development.
                        How to Understand Neural Coding?

                            Qinye Tong and Guang Li
            Department of Biomedical Engineering, Zhejiang University,
                          Hangzhou 310027, PR China

        Eric R Kandel put an end to the development of neuroscience over last one
hundred years [1]. He mentioned in the finis: Physicists and chemists have often
distinguished their disciplines from the field of biology, emphasizing that biology was
overly descriptive, atheoretical, and lacked the coherence of the physical sciences.
This is no longer quite true. By “This is no longer quite true”, Eric R Kandel referred
to the emergence of molecule and cellular neurobiology. Admittedly, analysis of
biochemical processes such as neurotransmitter, ion channel already brought us clear
results, yet it is useful to understand neural information. The change of molecule
element in neurons does influence the work of the whole neural system, just as the
deteriorated resistance will reduce the picture quality. However, the chemical
composition of the resistance and the picture are different concepts in two strata, in
other words, they are two things. Hence, even if the molecule and cellular
neuroscience is consummate one day, the main work of neuroinformatics will still be
qualitative description.

        Gilles Laurent put a paragraph in his article in the journal “SCIENCE”:
Studying a neural code requires asking specific questions, such as the following:
What information do the signals carry? What formats are used? Why are such formats
used? Although superficially unambiguous, such questions are charged with hidden
difficulties and biases.

        We believe that, in order to bring neuroinformatics from qualitative
description to rational analysis. To change “hidden difficulties and biases” in neural
coding, an essential problem (problem in conception) need to be solved. From the
ideas of traditional engineering technique, it is unbelievable that an unstable
instrument can precisely measure small signal. On the contrary, the olfactory
receptors in the dog‟s olfactory system finish their rebirth in 35 to 45 days. It is
amazing that such an unstable system could have outstanding sensitivity – the dog can
differentiate odors from several kilometers away. From the nonlinear viewpoint, the
traditional concepts about engineering technique are all linear which emphasize on
stable, balanced, orderly, deterministic and coincident aspects. But biology system is
nonlinear which emphasize on unstable, unbalanced, chaos, indeterminate and
inconsistent aspects. These two ideas are totally opposite to each other. To study the
mechanism of neural coding, we should discard linear views and learn to use
nonlinear concepts to analyze problems.

       Our opinion is neural information is processed in the stratum of neural
network composed of neuron connection, not in molecule level nor the neuron
ensemble level. A series of neural impulse represent the neural information. There are
underlying rules in this neuron induced impulses, which is determined by nonlinear
dynamics. Therefore, using nonlinear dynamics is the right way to decode neural
Several problems need to solve to verify that neuron impulses are the carrier of the

    1) Are there corresponding relationship between the neural impulse series and
       the input signals from outside?

    2) Is “impulse sequence space” an orderly space? If we gather all the neural
       impulses series in the neural system to form an “impulse series space” Ω, Ω
       must be an “orderly space”. This is because a very important feature of signal
       is the difference between big and small. Eyes could receive light signal that is
       bright or dark; ears could receive sound signal that is loud or feeble; it is the
       same that the olfactory system could receive odor signal that is strong or
       weak. To set up a corresponding relationship between the neural impulse
       series and the input signals from outside, we must ensure that Ω space must
       be a orderly space.

                Ω={ηi}                           ( 1)

       Ω is called sequence space. If we can elucidate Ω space, then
       neuroinformatics begin its time of rationality, otherwise it will keep in the
       level of qualitative description. What need to solve now is what
       characteristics does Ω space has. As proclaimed in the set theory, some sets of
       elements could form a group (as the renormalization group、 Galois group),
       some sets could form a ring of integers, or algebra (as the bool algebra).
       What system could Ω form? What kind of transform could be processed in Ω

     3) How to solve the conflict between the unstable neural system and the
       deterministic signal processing? Neural system is an extremely unstable
       system, there is the plasticity everywhere in the neural system. This is
       represented by the variability of ηi. Contradiction will happen when
       correspond each ηi to the deterministic information. Of course, the first idea
       appearing in head is the statistic method, in other words, considering the
       statistics of many ηi as the representation of a kind of information. Years of
       research prove that this method is useless. This is because the neural system is
       nonlinear and extremely unstable. The statistic method is useless in this
       situation. New ideas are needed to solve this problem.

        This paper gives an example of olfactory system to carefully analyze the
above four questions. First, sort the trajectories (impulse series), then define the
distance and transfer a problem from a ”trajectory space” into a ”distance space”.
Thus, according to the functional analysis, calculations are permitted in “distance

       The main task of studying neural coding is to explain how the irregular neural
impulse changes with the input signal. Upon finding this rule, we could elucidate the
information transform in neural system and the mechanism of neural coding. Circle
maps in neurons is an important theoretical tools.
        First, considering the response signal of a neuron under impulse stimulus of
invariable frequency. We use the classic H-H equation to describe potential change in
neurons (as in (1)). Though many people have improved the H-H equation, we know
that these improvements are only quantitative mending without qualitative
differences. Our discussion does not lose any generality.

The computer simulation result is listed below:
         I ext  g K n 4 V  E K   g Na m 3 hV  E Na   g l V  El 
         K t  An 1  n   Bn n 
         K t  Am 1  m   Bm m 
         K t  Ah 1  h   Bh h 
I ext  I offset  I sig
         T  6.3 
Kt  3            10

V '  V  Vrest
          0.01  10  V '
 An                  10 V ' 
               e               10       1
                                    V ' 80 
Bn  0.125  e
            0.1  25  V '
 Am                   25 V ' 
               e                   10
                          V '18 
Bm  4  e
                                        V '
 Ah  0.07  e                                   20

 Bh                  30 V ' 
              e                     10

According to the computer simulation, we get the figure of circle map:
        With this circle map rule, we can understand the meaning of irregular impulse
sequences. Taking the olfaction as the research object, we clarify the relation between
the thickness of the odor and the impulse series of all kinds of olfactory neurons, as
well as the rule that governs the information change between olfactory neurons.

        Raising the concept of “orderly space”, we admit the indetermination and
unstability in neural system. It is not necessary for all parameters in the H-H equation
to be precise and unchanged. Moreover, same conclusion can be drawn on the present
reformed the H-H equation if only the equation structure keep unchanged. In another
words, although the H-H equation describes the neurological electrophysiology
qualitatively our conclusion should be able to show the actual electrophysiology.
  A Computational Causality for Explanatory Language Understanding

                        Key-Sun Choi1 and Du-Seong Chang2
           Division of Computer Science, KAIST, BOLA, KORTERM, Korea
   Spoken Language Technology Division, KT, Division of Computer Science, KAIST,

        Causality or Causal relation refers to “the relation between a cause and its
effect or between regularly correlated events”. Causalities in the document are
represented implicitly, or by lexical patterns. Although the causality inference was
tried in some direction, the causality extraction and its applications are not sufficient.
In this paper, we focus on the extraction of causalities that are explicitly represented,
and we discuss some document understanding applications using the causality.

        We take prior consideration of the explicit causality that exists between noun
phrases. Causality patterns that connect two noun phrases are known to be causal
verbs. We use lexical patterns as a filter to find causality candidates and we transfer
the causality extraction problem to the binary classification. To solve the problem, we
introduce probabilities for word pair, concept pair and cue phrase that could be a
causality pattern. With these probabilities, we will propose a causality classifier based
on the Naïve Bayes classifier. The probabilities are learned from the raw corpus in an
unsupervised manner. With this probabilistic model, we increase both precision and
recall. Our causality extraction shows an F-score of 77.37%, which is an improvement
of 37.6% over the baseline model. The
long-distance causality is extracted
with the binary tree-styled cue phrase.

        When causality noun phrase
pairs are selected from one document
or from the same domain with the
same topic, we can make a causal
network like that shown in Figure 1.
When we assume that it is independent
from events that are not represented in
the graph, we can respond to some
causal questions such as “What is the
cause of the cancer?” „Cigarette
smoking‟ and „tobacco products‟ in the
sentence (1) are the causal information
of the term „gum disease‟. This causal         Figure 1: Causal network for „protein‟
information is one of pure collocation
information whose inter-term distance is so long that N-gram could not catch up.

        “Cigarette smoking and use of tobacco products may also cause gum
disease.”(1) Causality as a long-distance collocation information can improve the
performance of the text mining system. When we assume that two terms that have
similar causal information are similar terms, we can use the causal information for the
term classification. We prove that the causality is one of the positive features for the
term clustering. Our term clustering shows a precision of 70%, which is an
improvement of 32.9% over the lexical similarity-based model.
A new approach to localization and navigation of mobile robots - Effective
           Bayesian estimation and reinforcement learning.

                  Masumi Ishikawa, Fredrik Linåker, Keiji Kamei
                  Department of Brain Science and Engineering,
                      Kyushu Institute of Technology, Japan

Introduction: Two central issues in a mobile robot is how to make localization, i.e.
estimation of its location and orientation, and navigation to a given goal more
efficient. For efficient localization we pro- pose a new technique of using an omni-
directional camera and 35-dimensional rotation invariant feature vectors called polar
higher order local autocorrelation functions (PHLAC)[1].

        For efficient navigation we propose to introduce sensory information into
reinforcement learning (RL) to increase its learning speed and the probability of
reaching a goal, and to decrease the probability of collision [2]. We also propose to
use a genetic algorithm for optimizing the values of parameters in RL [3].

Localization: The PHLAC is an extension of HLAC, which is translation invariant.
Basic idea is that translation invariance on a panoramic view is equivalent to rotation
invariance on an omnidirectional view. It provides rotation invariance to the polar
HLAC directly on the omni- directional view. This is combined with particle filters,
which carry out 5000 localization hypotheses in parallel, resulting accurate estimation
of the location of a continuously moving real robot.

Navigation: A key idea of introducing sensory information is to directly modify value
functions (Q-values) near an obstacle based on sensory information in addition to the
modification of Q-values at the current location of a mobile robot based on a reward
from the environment in conventional RL.

        We also propose to optimize the values of parameters in RL with the help of a
genetic algorithm. An additional idea is to take advantage of inheritance in a genetic
algorithm; Q-values in RL in the previous generation are used as the initial Q- values
in the next generation to further accelerate learning.

Experimental results: The average location error is 31mm in the 1100mm by
800mm arena. Estimated body positions overlap the actual ones in over 95% of the
time steps. The average orientation error of the robot is 5.5 degrees.

        The average number of actions need- ed to reach a given goal for the
optimal values of parameters in RL decreases by about 30% compared
with RL with that for tentative parameters, resulting an approximately
shortest path. The number of goals reached increases more than 2 times
faster and the number of collisions is much smaller compared to
conventional RL.
Conclusions: Computer experiments have well demonstrated the effectiveness of
localization by PHLAC and of reinforcement learning by introducing sensory
information and a genetic algorithm.
References: [1] F. Linåker, M. Ishikawa, IROS-2004, pp. 4026-4031, Sendai, 2004.
[2] K. Kamei, M. Ishikawa, IJCNN, pp.3185-3188, Budapest, 2004.
       [3] K. Kamei, M. Ishikawa, A genetic approach to optimizing the values of
parameters in reinforcement learning for navigation of a mobile robot, to appear in
                             Neurobiology of Stress

                                Kyungjin Kim, Ph.D.
    Brain Research Center, 21st Century Frontier Program in Neuroscience, and
   School of Biological Sciences, Seoul National University, Seoul 151-742, Korea

        Stress in adulthood can exert profound influences on physiological and
behavioral consequences, but the extent to which prolonged maternal stress affects the
brain functions of adult offspring remained largely unknown. Chronic immobilization
stress to pregnant mice affected fetal development. When pups born from stressed
mice were reared in an environment identical to that of non-stressed normal condition,
several physiological parameters such as serum corticosterone level, body weight and
hippocampal mineralocorticoid (MR) and glucocorticoid (GR) mRNA levels in
maternally stressed offspring were similar to those shown in the control mice.
However, maternally stressed offspring showed a significant reduction in N-methyl-
D-aspartate (NMDA) receptor-mediated long-term potentiation (LTP) measured in the
CA1 area of hippocampal slices. A subsequent biochemical analysis indicated that
there was a clear decrease in synaptic NR1 and NR2B subunits of NMDA receptor in
the hippocampus with an apparent reduction of interaction between these two
subunits. Along with these electrophysiological and molecular aspects, Morris water
maze test and passive fear avoidance test showed that spatial learning and memory
and fear avoidance responses were impaired in maternally stressed adult offspring.
These results suggest that prolonged maternal stress leads to malfunctions of the
brain, which extend to and are revealed in adulthood.
     Can Stochastic Resonance Imaging be a substitute for high-priced
                       Gadolinium scan in India?

                      Prasun Roy, Vani K, T Ray, A K Saini
           Computational Neuroscience and Neuroimaging Laboratory
   National Brain Research Centre, Manesar, Gurgaon-122 050, Haryana, India

        The technique of using organometallic chelated compounds having high
magnetic moment as gadolinium and holmium („contrast agents‟ as gadodiamide)
have been an important armamentarium for neuroimaging, along with the novel
approach of nanoparticle based gadolinium targetting. However the cost of
gadolinium is expensive in the scenario of patients in developing countries. We probe
the possibility of using the approach of stochastic resonance imaging (SRI) to enhance
MR images so that the enhanced image could approximate gadolinium enhanced
images. Gadolinium enhancement of MR proton signal is basically effected by a
stochastic activation induced by gadolinium atoms. Proton relaxation processes occur,
actuated by microscopic effects as probabilistic fluctuation of local dipolar field due
to the stochastic kinetic motions of the gadolinium nuclei. We consider magnetization
of water medium as a dependent variable of the stochastic nature of nuclear relaxation
process, namely the noise intensity inducing the relaxation process that is effected by
the gadolinium; there is an optimal amount of gadolinium level needed for maximal

        We propose a bijective mapping between the magnetic signal intensity
continuum and the voxellated grayscale gamut and show transformationally that
stochastic activation of voxellated greyscale correlates with stochastic kinetic
gadolinium effect. We perform a stochastic resonance enhancement of the voxellated
MR signal and administer a programmed stochastic perturbation on pre-gadoliniated
T1 image of different brain lesions. There is appreciable enhancement of the images
which compare with the post-gadolinium images of the lesion. The noise correlogram
between stochastic resonant image and gadoliniated image shows an inverted U
graph, the characteristic signature of stochastic resonance, that indicates that
maximum enhancement and correlation occurs at a particular optimum stochastic
input level. We apply the technique to investigate MR image enhancement in
neoplastic and infective lesions of the brain (e.g. glioma and tuberculoma), and to
angiography. Future prospects could be to explore the possibility of evaluating SRI to
study „tropical ring lesions‟ of brain, a key neuroradiological challenge in India, being
hypothesized to be associated with chronic infective and inflammatory processes.
Regulation of Prefrontal Cortical Functions by Alpha-2-adrenoceptors: Its
     possible relevance to Attention Deficit Hyperactivity Disorder

                                     Bao-Ming LI
                     Institute of Neurobiology, Fudan University
                    220 Han-Dan Road, Shanghai 200433, China

        The prefrontal cortex (PFC) plays an essential role in the so-called executive
functions: the ability to hold information on-line in mind (working memory), regulate
our attention, inhibit inappropriate behaviors, and plan and organize for the future. It
is known that the PFC is very sensitive to change in the adrenergic environment.
Norepinephrine (NE) has marked effects on PFC functions, and these actions may
have particular relevance to Attention Deficit Hyperactivity Disorder (ADHD).

        During the past decade, our laboratory has focused on the alpha-2-adrenergic
regulation of prefrontal cortical functions. Blockade of alpha-2-adrenoceptors in the
dorsolateral PFC by local infusion of the alpha-2-adrenergic antagonist yohimbine
markedly impairs working memory ability in monkeys (Li et al. 1994). Conversely,
similar treatment with the alpha-2-adrenergic agonist guanfacine produces a delay-
dependent improvement in working memory (Mao et al. 1999). Consistently,
iontophoretically applied yohimbine suppresses PFC neuronal activity related to
working memory, whereas clonidine, either systemically or iontophoretically
administered, has an opposite effect (Li et al. 1999). Thus, alpha-2 adrenoceptors in
the PFC regulate working memory at both behavioral and cellular levels.

        Recently, we reported that some of the important symptoms of ADHD can be
recreated by blocking alpha-2 adrenoceptors in the PFC. In addition to the weakened
working-memory capability and working-memory related neuronal activity described
as above, infusions of yohimbine into the dorsolateral PFC increase impulsivity in
monkeys. During chronic intra-PFC administration of yohimbine, monkeys tested on
a go/no-go task show a selective deficit in no-go performance: they could not inhibit a
touching response to the no-go signal (Ma et al 2003). Meanwhile, the monkeys
demonstrate a dramatic increase in locomotor activity in home cages during treatment
with yohimbine (Ma et al 2004), very reminiscent of the increased activity found with
PFC ablations in monkeys or humans. Similar results have been repeated in rats with
yohimbine infusion into the medial prefrontal cortex (Ma et al, unpublished data).

        In addition, we also found that visuomotor associative learning, a task that
requires the ventral and orbital prefrontal cortex (Wang et al. 2000), is also sensitive
to manipulation of alpha-2 adrenoceptors in the PFC. Systemic administration or local
infusion of guanfacine into the ventral prefrontal cortex significantly enhances the
monkey‟s ability to map 1:1 associative relationships between visual patterns and
motor responses: the monkeys show an increased capability to apply win-stay/lose-
shift and change-stay/change-shift learning strategies (Wang et al. 2004a; Wang et al

       It has been speculated that striatal DA mechanisms underlie the hyperactivity
observed in ADHD patients. Our studies in monkeys and rats, together with studies by
other authors in this field, strongly suggest that, alpha-2-adrenergic activation tunes
the prefrontal cortex to an optimal functional status, and dysfunction of alpha-2 NE
system in the PFC may contribute to locomotor hyperactivity, impulsivity and poor
regulation of working memory, which form the cardinal symptomology of ADHD.
       Deciphering the genetic basis of mouse cerebellar development

                                Teiichi Furuichi
                    Laboratory for Molecular Neurogenesis,
               RIKEN Brain Science Institute, Wako 351-0198, Japan

        The brain is the ultimate genetic system to which a large number of genes are
devoted. In the post-sequencing era, it is now possible to elucidate how the brain
develops and functions on a genetic basis. We focus on the postnatal development of
the mouse cerebellum as a model system, since it develops through a series of
cytogenetic and morphogenetic events (cell proliferation and migration,
dendrogenesis and axogenesis, synaptogenesis, myelination, foliation and
fissurization, etc.) on schedule within the first three weeks of life. We attempt to
decipher the "genetic blueprint" for cerebellar development by profiling all of the
transcription (i.e., the transcriptome) responsible for developmental stages on a
genome-wide basis (e.g., fluorescence differential display, cDNA microarray, and
GeneChip). Then, the spatial (cellular) and temporal (developmental) specificities
associated with the expression patterns were analyzed by performing in situ
hybridization (ISH) brain histochemistry and reverse transcription-polymerase chain
reaction (RT-PCR), respectively. We have successfully systematized the information
collected regarding gene expression in an online "Cerebellar Development
Transcriptome (CDT) database". I will introduce about the transcriptomic feature of
cerebellar development and the CDT database. By applying these novel genome-wide
approaches, we are able to access a large number of differentially-expressed genes,
including those that had not yet been characterized in the nervous system, and those
which had only been predicted in silico. Among these novel cerebellar developmental
genes we identified, I will also introduce about the developmental roles of three
genes: Cupidin/Homer2 (a postsynaptic scaffold protein that interacts with

vesicle-associated protein that regulates the activity-dependent release of the
neurotrophins NT-3 and BDNF from the parallel fiber terminals of granule cells), and
Opalin (a transmembrane glycoprotein that specifically localizes to the paranodal loop
of the myelin sheath that wraps around the axons of Purkinje cells).
MAPK regulates phosphorylation of Neural Retina Leucine Zipper: A Key
    Regulator of Rod Photoreceptor Differentiation and Function

        Prabodh Swain, Sandeep Kumar, Dharmesh Patel, Anand Swaroop
   National Brain Research Centre, Manesar, Gurgaon- 122050, Haryana, India
  Department of Ophthalmology and Visual Sciences, University of Michigan, USA

         NRL is a bZIP group of DNA binding protein expressed in retina and pineal
gland. Mutation of NRL gene has been associated with different autosomal dominant
retinitis pigmentosa. One of the implications studied in the transactivation of
rhodopsin minimal promoter suggested that mutated NRL alters the synergistic
transactivation of rhodopsin promoter in the in vitro reporter assay. To understand the
exact biochemical alterations of NRL produced by such disease mutations we
expressed mutated proteins in vitro and performed in vitro phosphorylation.
Metabolic labelling of NRL revealed that S50T and P51L substitutions affected the
protein phosphorylation drastically. The multiple phosphoisoforms of NRL are
collapsed into 1-2 major protein bands in SDS-PAGE analysis. Further to understand
the nature of kinase(s) responsible for NRL phosphorylation, we identified a mitogen
activated protein kinase that specifically phosphorylates NRL. Deletion of the entire
transactivation domain of NRL abolished the MAPK mediated phosphorylation of the
protein. We suggest that MAPK is one of the kinase responsible for the
phosphorylation (primary post translational modification) of NRL. Such finding of the
role of MAPK in the phosphorylation of NRL provided an important clue to unravel
the complex regulatory processes involved in the rod differentiation and function.

Support:       Department of Biotechnology, Govt of India
 Examine synchrony of the relationship between blood pressure (BP) and
 renal sympathetic nerve activity (RSNA) in response to haemorrhage in
                               Wistar rats.
                            Tao Zhang1 and Zhuo Yang2
                                College of Life Sciences,
                 Medical School, Nankai University, PR China, 300071

       Recently we employed a power spectral technique and a cross-sample entropy
(CSE) method (Rickman & Moorman, 2000) to study synchrony of the relationship
between BP and RSNA during right atrial stretch to mimic plasma volume expansion
(Yang et al, 2002). CSE revealed that during the reflex inhibition there was more
synchrony between the oscillating signals in the BP and RSNA sequences. In the
present study we have used a similar analysis of these signals during a mild
haemorrhage which reflexly causes an increase in RSNA in an attempt to maintain BP

        The experiments were performed on 10 anaesthetised (urethane 650 mg.kg-1,
chloralose 50 mg.kg-1) Wistar rats. BP was measured from a femoral artery and
RSNA from a branch of renal nerve after exposing the left kidney retroperitoneally. A
33 second high frequency (1 KHz) sampling of BP and RSNA was recorded and
rectified. The trachea was cannulated and spontaneous respiration maintained. Rectal
temperature was maintained at 37oC by a heating blanket. A femoral vein was
cannulated and 1 ml of blood was removed into a pre-heparinised syringe over 1 min
and 5 mins later slowly reinfused. Data are expressed as mean ¡À S.E.M., and
analysed using repeated measures ANOVA. Statistical differences were considered
significant when p<0.05. Rats were killed by overdose of urethane at the end of

        Haemorrhage decreased BP and increased RSNA (25.9 ¡À 2.4%). A coherence
measurement from power spectral analysis failed to detect significant changes
between baseline and haemorrhage in either averaged coherence over the range 0-10
Hz or coherence at heart rate frequency. However a non-linear dynamic analysis of
the group data using CSE measurements showed that the relationship between BP
signals and RSNA time series did increase during haemorrhage. Such an increase in
entropy indicates that the volume reflex control system has a greater power to nullify
the disturbance than would be the case if increased nerve activity was more
synchronised. Intrathecal administration of the glutamate receptor antagonist,
kynurenic acid (2 mM) significantly reduced the reflex increase in RSNA (13.4 ¡À
2.9%) caused by haemorrhage. Analysis of the RSNA-BP time series during
kynurenic acid block showed there was no change in total power, power at heart rate
frequency, coherence at heart rate frequency, or in the cross-sample entropy
measurements. Thus the decrease in regularity between BP and RSNA during the
reflex increase in RSNA was prevented by blocking spinal glutamate receptor.

        The data from both linear and non-linear dynamic analysis together suggest
that reflex adjustments to blood volume disturbances are not mediated via brainstem
tone generating network oscillators but depend on enhancement of direct synaptic
input to final common pathway neurones.
Yang Z, Zhang, T & Coote JH (2002) Exp Physiol 87, 461-468
Rickman JS & Moorman JR (2002) Am J Physiol 278, H2039-H2049
   From Neural Firing to Hypersynchronous Discharge in the Cortex of
                          Epileptogenic Focus

                              Xin Tian & Yijun Song
                    The Department of Biomedical Engineering,
                 Tianjin Medical University, Tianjin, 300070, China

       The aim of this study is to investigate the hypersynchronous discharges in
epilepsy from neuron to brain with chronic temporal lobe epilepsy (TLE) rats.

       TLE rat model was prepared by lithium-pilocarpine method. The epilepsy –
like EEG in temporal lobe and hippocampus of TLE rats during seizures were shown,
which might be caused by hypersynchronous discharges. The neuronal apoptosis at
hippocampus were also detected in our study by in situ TUNEL.

       The question arisen here is the mechanism of the hypersynchronous discharges
in the Cortex related to hippocampus, although neurons decrease here due to the

        From the micro (neuron), there is a bursting in the firing neuron. There is a
high nonlinearity and nonstationarity of the bursting in neuron, and the bursting might
be consisted of several components with different frequencies. A time-frequency
coding is proposed and implemented to investigate the neurophysiologic mechanism
of the neural firing. The results of neural coding of firing neuron is applied to macro
to understand the hypersynchronous discharges happened in the cortex related with
hippocampus although there is the neuronal apoptosis.
             The Role of Ankyrins in Neurite Growth and Polarization

                               Hiroyuki Kamiguchi
                   Laboratory for Neuronal Growth Mechanisms
                      RIKEN Brain Science Institute, Japan

        During development, neurons send out neurites that differentiate into axons or
dendrites. Axons elongate and reach their appropriate targets to form neuronal
networks. These developmental events highly depend on various functional
molecules, including the cytoskeletons and cell adhesion molecules (CAMs). For
example, mutations of L1, a CAM in the immunoglobulin superfamily, cause
abnormal axon tract development in humans and mice. The cytoplasmic tail of L1
binds ankyrins that associate with the actin-spectrin cytoskeleton. The major ankyrin
isoforms expressed in the developing nervous system are 440-kD ankyrin-B and
480/270-kD ankyrin-G. Ankyrin-B mediates L1 coupling with retrograde actin flow,
thereby acting as a molecular clutch that regulates neurite growth. In contrast,
ankyrin-G and its associated protein,
maintains polarized L1 distribution in the axon. My talk will focus on the dynamic
and static roles of ankyrins in regulating neurite growth and axon-dendrite polarity.
        Blind Separation of Sound Sources in Real-World Situations

                      Kiyotoshi Matsuoka and Akio Yamazaski
                    Department of Brain Science and Engineering
                       Kyushu Institute of Technology, Japan

       Blind source separation (BSS) is a method for recovering a set of statistically
independent signals from the observation of their mixtures without any prior
knowledge about the mixing process. It has been receiving a great deal of attention
from various fields as a new signal processing technique. Among conceivable
applications of BSS the most promising one might be separation of sound signals.
Indeed, in the literature of BSS one can see a lot of papers that describe experiments
on sound separation. In our experience, however, although the conventional methods
for BSS are able to achieve separation for artificially synthesized data, they do not
necessarily work well for real-world data. Separation accuracy is often unsatisfactory
and, what is worse, they sometimes reveal incomprehensible instability.

        As known well, in the subject of BSS, the definition of the sources has a
certain indeterminacy: a source signal transformed by any linear filter can also be
considered a source signal. Corresponding to the indefiniteness of the sources, the
choice of the separator has an arbitrariness. For this point one of the authors proposed
a principle called the minimal distortion principle (MDP): “Among the feasible
separators, choose the one that best preserves the quality of the signals observed at the
sensors (microphones).” The BSS algorithm discussed in this paper also adopts the

        The task of ICA is basically to find the inverse of the mixing matrix and to
apply it to the observation. In practical applications the microphone array should be
made compact, but then the mixing matrix becomes almost singular particularly for
low frequencies. If the separator is constructed by a FIR filter with MDP for such a
nearly singular situation, a large number of taps or parameters (say, some thousands
taps) become necessary. It is not easy to determine such a large number of parameters
in a reliable manner. Actually we sometimes face the following phenomenon. When
applying an iterative ICA algorithm to a given data, in the beginning the algorithm
appears to behave in a desired manner, but as the iteration goes on, suddenly some
instability occurs.

        Looking at the result in such a situation, we usually found that the following
thing occurs. Some frequency components of a source appear at an output terminal of
the separator while other frequency components of the same source appear at a
different terminal. This phenomenon is probably due to a too high frequency
resolution when a long filter length is adopted. Time-domain algorithms for BSS are
usually thought to be relatively free from this kind of permutation problem as
compared to frequency-domain approaches, but it is not necessarily the case.
If adversely the algorithm is performed with a shorter filter length, the stability will
considerably be improved, but the accuracy of separation will be unsatisfactory. A
solution to this trade-off problem is to relax the MDP constraint. By relaxing the
constraint, we can reduce the length of the separating filter without degrading the
separation accuracy so much. Thus how to determine the degree of the normalizing
constraint, which is relevant to the quality of the separated signals, and the filter
length, is a very important problem. This paper shows some useful suggestions
indicating how to choose those parameters.
                     Neural Control of Saccade Sequences

                                  Supriya Ray
   National Brain Research Centre, Manesar, Gurgaon- 122050, Haryana, India

        The spatiotemporal organization of motor sequences enables complex goal
directed behaviors. We use the saccadic eye movement system as a simple model
system to understand the control of sequential behavior by having subjects perform a
variety of modified double-step tasks in which targets in different locations elicit
saccade sequences. By systematically varying the time at which the second target
appeared relative to the first saccade we show how sequential saccades maybe
programmed concurrently. Using different manipulations we describe results from
behavioral experiments in which sensory, attentional and cognitive influences
modulate the degree of parallel processing. We also show how such behavioral
experiments can be used in conjunction with electrophysiological experiments to test
how neural networks may implement behavioral control.
          Monitoring Ischemic Brain Injury Using Nonlinear Methods

                       Yisheng Zhua, Yihong Qiua, Shanbao Tongb
     Department of Biomedical Engineering, Medicine, Johns Hopkins University, USA
       Department of Biomedical Engineering, Shanghai Jiao Tong University, China

        The brain's electrical activity following: 1) graded ischemic injury; 2)
hypoxia-ischemic injury; 3) ischemic injury with pre-injection of NAALADae
inhibitor has been studied. Animal experimental models of brain injury were used.
Two channels of EEG, one channel of ECG were recorded continuously during the
experiments. EEG has been analyzed by Tsallis time-dependent entropy (TDE). Both
mean and variance of TDE have good specificity to injury and recovery. The mean
TDE decreases as the EEG becomes less complex during the brain injury; and TDE
increases gradually as the brain recovers. Heart rate variability (HRV) which were
extracted from ECG has been analyzed by several nonlinear algorithms. The brain
injury results in a reduction of HRV while a recovery brings HRV back to nearly
normal level. These results suggest that quantitative EEG and ECG can be used for
monitoring brain injury.
 Neuroanatomical Analysis for Onoamtopoeia and Phainomime Words:
                             fMRI Study

                         Jong-Hye Han and Kichun Nam
                 Department of Psychology, Korea University, Korea

         The purpose of this study is to examine the Neuroanatomical areas related
with onomatopoeia (sound-imitated word) and phainomime word (motion-imitated
word). Using the block-designed fMRI, whole-brain images (N=11) were acquired
during lexical decisions. We examined how the lexical information initiates brain
activation during visual word recognition. The onomatopoeic word recognition
activated the bilateral occipital lobes and superior mid-temporal-gyrus, whereas the
phainomime words recognition activated left SMA and bilateral cerebellum as well as
bilateral occipital lobes. Regions more activated for the phainomime word than
onomatopoeia included left SMA and bilateral cerebellum. Regions more activated
for the onomatopoeia than phainomime word included left superior and mid-temporal
gyri. The word recognition for onomatopoeia plus phainomime word showed
activation on bilateral middle and superior temporal gyrus, right supramarginal gyrus,
left middle temporal gyrus, left middle occipital gyrus, and right occipital gyrus. This
is the first fMRI research to analyze onomatopoeia and phainomime word.
                      Neuroinformatics and Data Sharing

           Ling Yin, Guang Li, Yiyuan Tang, Xianglan Jin, Xiaowei Tang
         Neuroinformatics Workgroup of China, 301 Hospital, Beijing, China

1. Development in OECD-GSF-NI-WG and Neuroinformatics

        In Jan 2004, there was a meeting of the Organisation for Economic Co-
operation and Development (OECD) committee for scientific and technological
policy at ministerial level. The 2002 Report on Neuroinformatics from the Global
Science Forum Neuroinformatics Working Group of OECD (OECD-GSF-NI-WG)
was submitted to the ministers attending the meeting. It highly recommended to
establish a new global mechanism, the INCF (International Neuroinformatics
Coordinating Facility), created an associated funding scheme, the PIN (Program in
International Neuroinformatics) and establish national nodes and research programs in
Neuroinformatics. It also recommended that the management and exploration of data
about the brain can be best achieved through a coordinated, multidisciplinary,
international effort. Ministers expressed their appreciation for the work of OECD-
GSF-NI-WG, and agreed that the study of human brain would be one of the most
difficult and rewarding scientific challenges of the 21st century. They also agreed that
interested countries should join together to create optimal conditions for the
expansion and international coordination in this new field. In Apr. 2004, the most
recent meeting on Neuroinformatics was held in Paris by OECD-GSF. Members from
21 countries attended that meeting. Details on several recommendations were warmly

2. Data Sharing in China

        Data sharing in Science and Technology is a very important policy in China.
Management and Sharing System of Scientific Data for Medicine is a key project of
the National Basic Platform for Science and Technology for year 2003 in the Ministry
of Science and Technology, P. R. China. As the paramount part of the National
Engineering of Scientific Data Sharing, the System is undertaken by Chinese
Academy of Medical Sciences, Chinese Center for Disease Prevention and Control,
Chinese PLA General Hospital, and Chinese Academy of Traditional Chinese
Medicine. Scientific data resources in medicine will be integrated together in the way
of distribution physically and unification in logic by the system. The system covers
most of the fields in medicine, including basic medicine, clinical medicine, public
health, traditional Chinese medicine, special medicine, pharmacology and innovated
drug etc. In addition, the system also specially sets up database for SARS and
respiratory diseases.

3. Some Suggestions to Neuroinformatics and Data Sharing

        From the above, we can see there are some overlaps between
Neuroinformatics and Data sharing. In a way, Neuroinformatics is data sharing in all
levels of neuroscience. Some experiences and suggestions about data sharing are also
suitable to Neuroinformatics, as shown the following: 1) In order to strengthen
international cooperation of establishment of Neuroinformatics data bank we need to
set up unified data bank criteria on the basis of data integration and modification from
scattered scientists and facilities. We need to set up bank groups of neuroscience data,
something similar to gene bank to make it sharable, extendable to reach some kind of
scale and authorization. 2) In order to reach true meanings of data sharing, we need to
establish an integrated service system of international data management and sharing
of neuroscience data to break the barriers among basic medicine, clinic medicine,
preventive medicine, public health and pharmacology, etc. It will offer one-stop
service via index and catalogue inquiries for data service and network environment of
therapeutics, prevention, control and research of specific diseases. 3) To speed up
collecting top human resources for access and sharing of neuroscience data, normal
training programs should be offered.
  Improvement of reinforcement learning of a mobile robot using sensors
                        and a genetic algorithm

                         Keiji Kamei and Masumi Ishikawa
                       Dept. of Brain Science and Engineering,
                       Kyushu Institute of Technology, Japan

Background and Idea: Reinforcement learning (RL) is frequently used in mobile
robots, but suffers from slow learning. To solve this difficulty we propose to intro-
duce sensory signals into RL in a direct way. Another difficulty is that we generally
don‟t have prior information on the values of parameters in RL. To solve this
difficulty we propose to use a genetic algorithm (GA) with inheritance for optimizing
the values of parameters in RL.

Methods: In addition to directly modify value functions (Q-values) near an obstacle
based on sensory information, we also restrict the reduction of Q-values to only the
region where the distance to the obstacle is less than a given threshold. The latter
helps to generate the shortest path to a given goal and to pass through a narrow

        RL has such parameters as a discount rate, a learning rate, an  in the  -
greedy policy, rewards for actions, a reward from the environment, a reward from
sensory information, and its thresh- old in modifying Q-values. The length of a
chromosome is 48 bits, with 6 bits for each parameter. Parameters are coded in a
linear scale except for a discount rate, which is coded in a logarithmic scale.

       Let the probability of the selection of each gene locus and crossover be 10%.
50 individuals are generated initially, for each of which fitness is evaluated. We then
generate 25 new individuals in addition to the original 50 individuals. Out of 75
individuals, 50 individuals with higher fitness are selected. The resulting 50
individuals constitute the next generation.

Results of Computer Experiments: In computer experiments using a mobile robot
in Fig. 1 we assume 3 kinds of actions, i.e., moving forward by 10cm, turning right or
left by 10 . It has ultrasonic sensors, which can accurately measure the distance to an
obstacle not exceeding 80cm. Table 1 indicates the results of computer experiments.
Fig.1 Mobile robot “TRIPTERS mini”, (a) overview, (b) positions of sensors.

Table 1 The results of computer experiments over the last 500 episodes. “Optimal”
and “tentative” stand for the performance by the optimal parameters
and by tentative parameters, respectively.
                                 #forward #rotation   #goals
                     optimal       24.31    19.95      500
                     tentative     39.16    23.09      500

Conclusions: Our key idea is to optimize the values of parameters in RL with the
help of a GA with inheritance. We demonstrate that the optimization of parameters
decreases the average number of for- ward actions and that of turning by about 40%
and 20%, respectively. Moreover, all the episodes succeed in reaching a given goal.
      Robustness, Evolvability, and Optimality of Evolutionary Neural

                            Paulito Palmes , Shiro Usui
             Neuroinformatics Laboratory RIKEN Brain Science Institute
                 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan

        Artificial neural network (ANN) is a popular tool of researchers for activities
that involve machine-learning tasks such as prediction, generalization, and
classification. One of the major problems for its successful implementation, however,
is the absence of a general rule to guide implementers on how to choose the
appropriate architecture and initial state in a given problem domain. Since majority of
ANN approaches rely on the gradient approach to discover the correct relationship
between its input and output, inappropriate choices of its architecture and initial states
often lead to overfitness or underfitness of the training data. One promising approach
to help in the ANN design problem is the use of evolutionary computation (EC).
ANN design can be regarded as an optimization problem where the goal is to evolve
optimal network structure and weights. There are two major ways to carry out this
evolution, namely: invasive evolution and non-invasive evolution. Non-invasive
approach evolves ANN structure by EC and adapts its weights by BP
(backpropagation). Since weights are not stored during evolution, its structure
evolution forces retraining of all its weights by BP which makes the entire process
inefficient and prone to the local optima problem. On the other hand, invasive
approach simultaneously evolves ANN‟s structure and weights without relying on the
gradient information. Since both weights and structures are locally stored and
participate in the evolution process, optimal weights and structures are not relearned
but propagated to the succeeding generations. These simultaneous evolutions create a
feedback loop such that any shortcoming in its structure exploration is compensated
by its weight adaptation, and vice-versa. Our empirical investigation shows that
invasive evolution is robust from various forms of perturbation and produces ANN
with stable and optimal solution.
    Korean Sentence Processing Mechanisms reflected on ERP patterns

                     Choong-Myung Kim, and Kichun Nam
                Department of Psychology, Korea University, Korea

        Korean language has different structures from Indo-European languages and
other Asian languages. The current study was designed to examine the ERP patterns
and the cortical areas related with the lexical, syntactic, and semantic component in
Korean sentence comprehension. In Experiment 1, ERP patterns related with the
reading comprehension were examined and in Experiment 2, ERP patterns associated
with the listening comprehension were investigated. For the syntactic violation, P600
and Late Positivity after 700 msec were measured and for the semantic anomaly,
N400 and Late Positivity were observed in Experiment 1 and 2 commonly. P600
components for the syntactic violation occurred in the left frontal and central lobes,
while N400 components for the semantic anomaly appeared in the left all lobes. The
difference between reading and listening comprehension is that the N100 component
is measured for the semantic anomaly in listening, whereas the N100 component is
recorded for the syntactic violation in reading. These results implicate that the early
sentence processing varies depending on the sentence presentation modality, while the
semantic and syntactic processing is common across modalities.
        Signal transduction of auto-regulatory microglial apoptosis

           Kyoungho Suk, Heasuk Lee, Eunyung Son, Jaeyoon Jeong,
                  Jayoung Lee, Boyoung Jung, Dae Young Jung
 Department of Pharmacology, Kyungpook National University School of Medicine,
                            Daegu, 700-422, Korea

        Activation-induced cell death (AICD) is an auto-regulatory mechanism for the
immune system to remove unwanted activated immune cells after making appropriate
use of them. Although AICD has been first identified in lymphocytes, recent works
indicated that both microglial cells and astrocytes in CNS might be under the control
of a similar regulatory mechanism. In contrast to AICD of T lymphocytes where Fas-
FasL interaction plays a central role, neither Fas-FasL interaction nor TNF was
important in AICD of microglial cells. Instead, nitric oxide (NO) produced by
activated microglial cells themselves was the major cytotoxic mediator.
Inflammatory stimuli such as LPS and IFN played a multiple role in AICD of
microglial cells. They not only induced the indirect apoptotic pathway via production
of NO, but also initiated the direct apoptotic pathway through the induction of
caspase-11 or antiproliferative BTG1 gene. While caspase-11 induction and its
activation were required for NO-independent apoptotic pathway, IRF-1 and NF-B
were involved in NO-dependent apoptosis of microglial cells mainly by mediating
NO synthesis (via iNOS induction). BTG1 participated in the AICD of microglia by
lowering the threshold for apoptosis; BTG1 increased the sensitivity of microglia to
apoptogenic action of NO. The auto-regulatory apoptosis of activated microglia was
mediated through Toll-like receptor (TLR)4, but not TLR2, based on the studies using
TLR2 or TLR4-deficient mice and dominant negative mutants. The main difference
between TLR2 and TLR4 signaling in microglia was IRF-3 activation and IFN
production followed by STAT1 activation; while TLR4 agonist induced IRF-3
activation and IFN production, TLR2 did not. Taken together, AICD of microglia
appears be an auto-regulatory mechanism that controls the microglial activation, and
the failure of the auto-regulatory mechanism may be in part responsible for the
deleterious effects of microglial activation associated with CNS pathologies. Thus,
further elucidation of molecular mechanisms underlying the auto-regulation of
microglial activation may enhance our understanding of pathogenesis of CNS
disorders such as neurodegenerative diseases.
Informal Inference based on the Integration of Multiple Neural Networks

                      Kyung-Joong Kim and Sung-Bae Cho
               Department of Computer Science, Yonsei University,
             134 Shinchon-dong, Sudaemoon-ku, Seoul 120-749, Korea

        In human brain, the interaction of multiple modules generates a high-level
functionality such as inference and cognition and incorporates both symbolic and
connectionist processing. As inference is least developed and unknown function of the
brain, it is a challenging problem to explore and implement the function in order to
discover the brain and apply it to several interesting problems. Usually, inference of
human brain is not static but dynamic and informal. Implementing such characteristics
is one of the challenging tasks for cognitive modeling. Similar to the human brain, the
integration of multiple models for better performance can be exploited as an inference
model. In this research, we attempt to devise a number of neural network models and
combine them using behavior network and fuzzy integral which have flexibility for
developing complex adaptive systems and apply the idea into real-world applications
including web usage prediction and robot control.

        Since exponentially growing web contains giga-bytes of web documents, users
are faced with difficulty to find an appropriate web site. Using profile, information
retrieval system can personalize browsing of the web by recommending suitable web
sites. User's evaluation on the web contents can be used to predict users‟ preference
on web sites and construct profiles automatically. User profile represents different
aspects of user's characteristics, thereby we need an ensemble of neural networks that
estimate user's preference using the web contents labeled by user as "like" or
"dislike." Fuzzy integral is a combination scheme that uses subjectively defined
relevance of models and structure adaptive self-organizing map (SASOM) is a variant
of SOM that is useful to pattern recognition and visualization. Fuzzy integral-based
ensemble of SASOM‟s trained independently is used to estimate user profile and
tested on UCI Syskill & Webert data. Experimental results show that the proposed
method can perform better than not only previous naïve Bayes classifier but also
majority voting of SASOM‟s.

       Similarly, NN (Neural Network) based on CA (Cellular Automata) models
complex phenomenon by simple rules, and optimized by genetic algorithm. Like
many evolutionary approaches to robot control such as neural network evolved by
genetic algorithm and fuzzy controller optimized by genetic algorithm, NN based on
CA can be applied to robot control. Behavior modules such as avoiding obstacles and
following light are evolved on the model. They are evolved incrementally by starting
with simpler environment needed simple behavior and gradually making it more
complex and general for complex behaviors. Because evolving higher behaviors
directly is difficult, we combine several basic behaviors by symbolic behavior
network. Robot selects one of the basic behavior modules evolved or programmed at
each time. We evaluate the performance of robot using Khepera simulator and modify
simulator interface for visualization of the action selection procedure. Simulation
results show the possibility of the symbolic combination of action modules for higher
         Unsupervised Extraction of Video Features for Lipreading

                    Michelle Jeungeun Lee1 and Soo-Young Lee1,2
                               Department of BioSystems
                Brain Science Research Center and Department of EECS
             Korea Advanced Institute of Science and Technology, Korea

        It is very important to know the basic components of the patterns we want to
recognize or synthesize. For speeches the basic components, also called as features,
are Gabor-like signals localized both in time and frequency. For images the features
are Gabor-like edge patterns. In this paper we report the features of video streams
extracted from the video clips of human lip motion. Three unsupervised algorithms
are investigated for the extraction of the video features. The Principal Component
Analysis (PCA) results in global features, while the Independent Component Analysis
(ICA) and Non-negative Matrix Factorization (NMF) result in local features. The
effects of the time frames are analyzed and their statistical characteristics are
investigated. The resulting features may be applicable to lip-reading and generation of
lip videos for specific sounds.
 Glutamate receptor-mediated signaling and the functional implication in
                           microglial cells

                             Su-Yong Eun, M.D., Ph.D.
                 Div. of Brain Disease, Dept. of Biomedical Science
                    National Institute of Health 122-701, Korea

        It has been recently shown that the expression of various types of
neurotransmitter receptors are not restricted to neurons but also observed in majority
of glial cells. However, their function in glial cells is not known well in both
physiological and pathological conditions. Here, we investigated the role of glutamate
receptor on immediate-early genes (IEGs) in primary cultured and BV-2 microglia.

        Our results demonstrated that both c-fos and c-jun mRNA and protein were
dramatically induced following treatment with various glutamate receptor agonists
(500 μM); N-methyl-D-aspartic acid (NMDA), kainic acid (KA), (S)-α-amino-3-
hydroxy-5-methyl-4-isoxazolepropionic        acid      (AMPA)       and     (RS)-3,5-
dihydroxyphenylglycine (DHPG). The responses were significantly suppressed by
specific antagonists and also by calcium chelating agents EGTA and BAPTA-AM.
Our results suggest that glutamate receptor activation regulates IEGs gene expression
by modifying intracellular calcium levels in microglia. In addition, glutamate
treatment markedly induced microglial cell proliferation, morphological
transformation and the expression of inflammatory cytokines such as inducible nitric
oxide synthase (iNOS) and tumor necrosis factor (TNF-alpha). Our results present the
evidences that glutamate might be a potent microglial activator in a certain
pathological condition. These findings might provide an insight to understand the
function of microglial glutamate receptors in neuron-to-glial interaction under the
excitotoxic conditions.
    Programmed cell death of adult generated hippocampal neurons is
              mediated by the pro-apoptotic gene Bax

                                 Woong Sun
  Department of Anatomy, Brain Korea 21, Korea University College of Medicine,
             126-1 Anam-Dong, Sungbuk-Gu, Seoul 136-705, Korea

        In the dentate gyrus (DG) of the adult mouse hippocampus, a substantial
number of new cells are generated daily, but only a subset of these survive and
differentiate into mature neurons, whereas the majority undergo programmed cell
death (PCD). However, neither the intracellular machinery required for adult stem
cell-derived neuronal death, nor the biological implications of the significant loss of
these newly generated cells have been examined. Several markers for apoptosis failed
to reveal cell death in Bax-deficient mice and this, together with a progressive
increase in neuron number in the DG of the Bax-KO, indicates that Bax is critical for
the PCD of adult-generated hippocampal neurons. Whereas the proliferation of neural
progenitor cells was not altered in the Bax-KO, there was an accumulation of
doublecortin (DCX), calretinin (CR)+ and NeuN+ postmitotic neurons, suggesting
that Bax-mediated PCD of adult-generated neurons takes place during an early phase
of differentiation. The absence of PCD in the adult also influenced the migration and
maturation of adult generated DG neurons. These results suggest that PCD in the adult
brain plays a significant role in the regulation of multiple aspects of adult
 Biotransformation of drugs mediated by brain-specific splice variants of
            the drug-metabolizing enzyme, Cytochrome P450

    Reddy P. Kommaddi, Harish V. Pai, Shankar J. Chinta and Vijayalakshmi
                Division of Molecular and Cellular Neuroscience,
   National Brain Research Centre, Manesar, Gurgaon- 122050, Haryana, India

        Cytochromes P450 (P450) is a family of heme proteins that functions as
mono-oxygenase and metabolizes a variety of xenobiotics, including drugs. Liver is
quantitatively the major organ involved in P450 mediated metabolism. However,
presence of P450 and associated mono-oxygenase activities in extrahepatic tissues
such as lung, kidney and brain has been demonstrated. P450-mediated metabolism of
psychoactive drugs in brain leads to pharmacological modulation at site of action and
results in variable drug response. A frame-shift mutation 138delT generated an open
reading frame in the pseudogene, CYP2D7 and an alternate spliced functional
transcript of CYP2D7 containing partial inclusion of intron 6 was identified in human
brain but not in liver or kidney from the same individual. The mRNA and protein of
the brain variant CYP2D7 was detected in 6 out of 12 human autopsy brains.
Genotyping revealed the presence of the frame-shift mutation 138delT only in those
human subjects who expressed the brain variant CYP2D7. Thus, genotyping revealed
the presence of frame shift mutation 138delT only in those human subjects who
expressed the brain variant CYP2D7, which demethylates codeine to morphine.
CYP1A1, a P450 enzyme bioactivates polycyclic aromatic hydrocarbons to reactive
metabolites, which bind to DNA and initiate carcinogenesis. RT-PCR analyses using
autopsy human brain samples demonstrated the presence of a splice variant of
CYP1A1 in human brain with exon-6 deletion. This splice variant present in all 23
human brain samples that were examined could potentially biotransform drugs by
pathways that are different from the wild type enzyme. We demonstrate the presence
of unique P450 enzymes in human brain that are generated by alternate splicing and
mediate biotransformation reactions that are dissimilar from known pathways in liver.

To top