This action might not be possible to undo. Are you sure you want to continue?
The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright
Author's personal copy
ARTICLE IN PRESS
Neurocomputing 71 (2008) 3367–3371
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/neucom
Network properties of a model for conscious and unconscious mental processes$
´ Roseli S. Wedemann a,Ã, Luıs Alfredo Vidal de Carvalho b, Raul Donangelo c
a b c
´ ´ ˜ Instituto de Matematica e Estatıstica, Universidade do Estado do Rio de Janeiro, R. Sao Francisco Xavier, 524, 20550-013 Rio de Janeiro, Brazil Programa de Eng. Sistemas e Computac ˜o - COPPE, Universidade Federal do Rio de Janeiro, Caixa Postal 68511, 21945-970 Rio de Janeiro, RJ, Brazil -a ´ Instituto de Fısica, Universidade Federal do Rio de Janeiro, Caixa Postal 68528, 21941-972 Rio de Janeiro, RJ, Brazil
a r t i c l e in fo
Available online 15 July 2008 Keywords: Consciousness and unconsciousness Neuroses Hierarchical memory Self-organized learning Complex neural network properties
We have previously described the mental pathology known as neurosis, in terms of its relation to memory function. We proposed neural network mechanisms, whereby neurotic behavior is described as a brain associative memory process, and the symbolic associativity involved in psychoanalytic workingthrough can be mapped onto a corresponding network reconﬁguration process. Microscopic mechanisms that control synaptic properties self-organize the memory networks to a hierarchical, clustered structure. Modules corresponding to sensorial and symbolic memories interact, representing unconscious and conscious mental processes. Here, we review these concepts and illustrate, with simulations, some of these complex networks’ behaviors and properties. & 2008 Elsevier B.V. All rights reserved.
1. Introduction Although inconclusive, psychodynamic theories [7,9,10,17] seem to suggest correlations between creativity, associativity, psychopathology and the unconscious. We explored these commonalities and proposed, in a previous paper , a schematic model for some concepts related with neurotic mental processes, as described by Freud [7,9,10]. Our description is based on the current view that the brain is a cognitive system composed of neurons, interconnected by a network of synapses, that cooperate among themselves to process information in a distributed fashion. Mental states are thus the result of the global cooperation of the brain’s distributed neural activity [5,13,14]. The emergence of a global state of the brain’s neural network generates a bodily response, which we call an act. Psychoanalytic research regarding the transference neuroses has found that traumatic and repressed memories are knowledge which is present in the subject, but which is symbolically inaccessible to him. It is therefore considered unconscious knowledge [7,9]. Freud observed that neurotic patients systematically repeated symptoms in the form of ideas and impulses, and called this tendency a compulsion to repeat . He related the compulsion to repeat to repressed or traumatic memory traces, caused by a mental conﬂict . Neurotic analysands have been
$ This research was developed with grants from the Brazilian National Research Council (CNPq), the Rio de Janeiro State Research Foundation (FAPERJ) and the Brazilian agency which funds graduate studies (CAPES). Ã Corresponding author. Tel.: +55 2122382180; fax: +55 2125877212. E-mail addresses: firstname.lastname@example.org (R.S. Wedemann), LuisAlfredo@ufrj.br (L.A.V. de Carvalho), email@example.com (R. Donangelo).
able to obtain relief and cure of painful symptoms through a mechanism called working-through. This procedure aims at developing knowledge regarding the causes of symptoms by accessing unconscious memories, and understanding and changing the analysand’s compulsion to repeat . The technique involves mainly free associative talking during analytic sessions and interpretation of dreams, among other processes. In , memory was modeled by a Boltzmann machine (BM) represented by a complete graph. It is known, however, that brain neuronal topology is selectively structured. Neurons interact mainly with spatially close neighbors, having fewer long-range synaptic connections to more distant neighbors [5,12,14]. We further developed the memory model by including some known microscopic mechanisms that control synaptic properties, so that the network self-organizes to a hierarchical, clustered structure. We propose an organization, where two hierarchically structured modules corresponding to sensorial and symbolic memories interact, producing sensorial and symbolic activity, representing unconscious and conscious mental processes. In proposing this organization, we followed Freud’s idea that unconscious memories are those which we cannot access symbolically, i.e. cannot talk about [7–10]. In this paper, we represent brain mechanisms involved in neurosis as a complex system, and analyze them according to recent methods developed for the study of complex networks. A review of recent developments in the scientiﬁc understanding of consciousness, as well as a model for attention as a basic function related to conscious activity may be found in . Kinsbourne  discusses how Freud’s attempt at proposing a neural substrate for mental processes  can be viewed in light of modern developments in neuroscience, such as the understanding
0925-2312/$ - see front matter & 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.neucom.2008.02.023
Author's personal copy
ARTICLE IN PRESS
3368 R.S. Wedemann et al. / Neurocomputing 71 (2008) 3367–3371
of forebrain functioning. Aleksander and Morton review their Axiomatic Consciousness Theory (ACT) in , which addresses phenomenology in neuroscience (relating symbolic representation to subjective experience), and apply it to modeling of visual phenomenology. The Global Workspace Theory  is based on the computational science concept of global workspace, where material needed to be worked on by a number of processors is held, and gives an useful view on certain aspects of consciousness. Our main contribution with respect to current work regarding machine models of consciousness is to propose a neuronal mechanism that describes conscious and unconscious memory activity involved in neurosis. The unconscious compulsion to repeat is explained as a neural network associative memory mechanism, where an input stimulus of any kind associates to a pattern in sensorial memory, which cannot activate symbolic brain processing areas. Neurotic (unconscious) acts are isolated from symbolic representation and association (similar to reﬂexes). We illustrate by our network model, how Freud’s ideas regarding the unconscious show that symbolic processing, language and meaning are essential for consciousness. We ﬁrst review our model for neurosis  and the algorithm for generating hierarchically clustered memory modules . We then present new simulation results and model analysis, with some important properties of these complex networks. Finally, we draw conclusions and perspectives for future work.
3. Self-organizing hierarchical memory structures As a ﬁrst approximation , we modeled memory by a BM represented by a complete graph. In a further reﬁnement, we now consider that neurons belong to two different subsets, corresponding to sensorial and symbolic memories. Memory traces stored in sensorial memory represent mental images of stimuli received by sensory receptors (located in eyes, ears, skin and other parts of the body) from the environment and the body itself. This includes information regarding affects and emotion. Sensorial memory represents brain structures such as the amygdala, cerebellum, reﬂex pathways, hippocampus, and prefrontal, limbic and parieto-occipital-temporal cortices, which synthesize visual, auditory and somatic information. Symbolic memory stores representations of traces in sensorial memory, i.e. symbols, and refers to another level of representation. It represents brain structures such as areas of the medial temporal lobe, the hippocampus, Wernicke’s and Broca’s areas and other areas of the frontal cortex. These latter areas are associated with language and, because of them, we can associate a word such as ‘‘red’’ to the sensation of seeing a red object. Attentional mechanisms , which we did not model, select stimuli to sensorial memory and allow them to become conscious, if associated to a symbolic memory trace. Sensorial stimuli which cannot associate to symbolic memory remains unconscious [7,9,14]. We refer to the subset of memory traces in sensorial memory which can associate to a symbol as the ‘‘conscious’’ subset. Freud called this the system conscious and deﬁnes it as the set of psychical information which is capable of becoming conscious (in some of his works he refers to this set as the preconscious) [7,9]. The neurotic compulsion to repeat [9,10] is explained here as a bodily response (an act) to an access to sensorial memory, which does not activate symbolic memory, as in a reﬂex. This accounts for the fact that neurotics say they cannot explain their neurotic acts, such as in ‘‘I don’t know why I was so upset and overreacted with Mary’’. When a stimulus S 0 which causes the retrieval of a sensorial memory trace can activate also a pattern in symbolic memory, it can become conscious, the output is not as in reﬂexive behavior and there is another level of processing. This mechanism is similar to ideas proposed by Edelman  and strongly reﬂects Freud’s concepts of conscious and unconscious mental processes and the role of language in psychoanalysis [9,10]. In order to model the organization of the topology of each memory, we consider the following microscopic biological mechanisms. Brain cells in many animals have a structure called on-center/off-surround, in which a neuron is in cooperation, through excitatory synapses, with other neurons in its immediate neighborhood, whereas it is in competition with neurons that lay outside these surroundings. These mechanisms are found statically hardwired, and also as part of dynamical processes, where neurons compete for certain chemicals [12,14]. In synaptogenesis, for example, substances generically called neural growth factors are released by stimulated neurons and, spreading through diffusion, reach neighboring cells, promoting synaptic growth. Cells that receive neural growth factors make synapses and live, while cells that have no contact with these substances die . A neuron releasing neural growth factor guides the synaptic formation in its tri-dimensional neighborhood, becoming a center of synaptic convergence. When neighboring neurons release different neural growth factors in different amounts, many synaptic convergence centers are generated and a competition is established between them, through the surrounding synapses. A signaling network triggered by environmental stimulation is established, which controls development and plasticity of neuronal circuits, and this suggests an idea of the way environment represents itself in the brain. We thus developed the following
2. Functional and computational model for the neuroses We review the model described in previous work , where we proposed that the neuroses manifest themselves as an associative memory process, a mechanism where the network returns a stored pattern, when it is shown another input pattern similar to the stored one . We modeled the compulsion to repeat neurotic symptoms, by supposing that such a symptom is acted, when the subject is presented with a stimulus, which resembles a repressed or traumatic (unconscious) memory trace. The stimulus causes a stabilization of the neural net onto a minimal energy state, corresponding to the memory trace that synthesizes the original repressed experience, which in turn generates a neurotic response (an act). The neurotic act is not a result of the stimulus as a new situation but a response to the repressed memory. We mapped the linguistic, symbolic associative process involved in psychoanalytic working-through into a corresponding process of reinforcing synapses among memory traces in the brain. These connections should involve symbolic memory, leading to at least partial transformation of repressed memory to consciousness. This has a relation to the importance of language in psychoanalysis and the idea that unconscious memories are those that cannot be expressed symbolically. We propose that as the analysand symbolically elaborates manifestations of unconscious material through transference in psychoanalytic sessions, he is reconﬁguring the topology of his neural net, by creating new connections and reinforcing or inhibiting older ones. The network topology which results from this reconﬁguration will stabilize onto new energy minima, associated with new acts. The process of slowly and repeatedly reconﬁguring synaptic connections to elaborate knowledge accounts for the long durations of psychoanalytic processes, where repetition is specially important. Memory functioning is modeled by a BM with N neurons, where node states take binary values. Pattern retrieval on the net is achieved by a standard simulated annealing process, in which the network temperature T is gradually lowered by a factor a. We refer the reader to Ref.  for a description of the BM algorithm and its properties.
Author's personal copy
ARTICLE IN PRESS
R.S. Wedemann et al. / Neurocomputing 71 (2008) 3367–3371 3369
clustering algorithm, to model the self-organizing process which results in a structured topology of each memory. Step 1: Neurons are uniformly distributed in a square bidimensional sheet. Step 2: To avoid the unnecessary and time-consuming numerical solution of the diffusion equation of the neural growth factors, we assume a Gaussian solution. Therefore, a synapse is allocated to connect a neuron ni to a neuron nj according to a ~ ~ ~ Gaussian probability, given by P ij / expðÀðr j À r i Þ2 =2s2 Þ, where r j ~ and r i are the positions of nj and ni , respectively, in the bi-dimensional sheet and s is the standard deviation of the distribution, which is considered here a model parameter. If a synapse is allocated to connect ni and nj , its strength wij is proportional to Pij . Synaptic weights are symmetrical, wij ¼ wji . Step 3: We veriﬁed in previous work [4,21] that cortical maps representing different stimuli are formed, such that each stimulus activates a group of neurons spatially close to each other, and that these groups are uniformly distributed along the memory sheet. We thus randomly choose m neurons which will each be a center of the representation of a stimulus. The value of m is chosen considering the storage capacity of the BM . Step 4: For each of the m centers chosen in Step 3, reinforce adjacent synapses according to the following criteria. If ni is a P center, deﬁne sumni ¼ j jwij j, where wij is the weight of the synapse connecting nj to ni . For each nj adjacent to ni , increase jwij j by Dwij, with probability Probnj ¼ jwij j=sumni , where Dwij ¼ Z Probnj and Z 2 R is a model parameter chosen in ½0; 1. After incrementing jwij j, decrement Dwij from the weights of all the other neighbors of ni , according to: 8kaj; jwik j ¼ jwik j À Dwik , P where Dwik ¼ ð1 À jwik j= kaj jwik jÞDwij . Step 5: Repeat Step 4 until a clustering criterion is met. In the above clustering algorithm, Steps 2 and 3 are justiﬁed in the algorithm’s description. Step 4 regulates synaptic intensities by strengthening synapses within a cluster and reducing synaptic strength between clusters (disconnects clusters). Neurons that have received stronger sensorial stimulation, and are therefore more strongly connected, will stimulate their neighborhoods and promote still stronger connections. The growth of long-range synapses is energetically more costly than short-range synaptic growth, and therefore the former are less frequent in the brain than the latter. For allocating long-range synapses which connect clusters, we should consider the basic learning mechanism proposed by Hebb [5,13,14] based on the fact that synaptic growth among two neurons is promoted by simultaneous stimulation of the pair. Neuronal groups whose states store certain memory traces, which receive simultaneous stimuli, should enhance synaptic growth among the neuronal groups representing these traces, allowing association among traces. Since memory traces represent both sensorial information and concepts (symbolic memory), we are also representing association of ideas or symbols by long-range synapses. This suggests a mechanism through which the external world, culture and basic language structure may be mapped onto brain topology. Since we are still not aware of the synaptic distributions that result in such topologies, as a ﬁrst approximation, we allocate long-range synapses randomly among clusters. Within a cluster C, ni is chosen for the connection with probability P P P P i ¼ j jwij j= nj 2C k jwjk j. If the synapse connects clusters in different memories (sensorial and symbolic), its randomly chosen weight is multiplied by a factor k 2 R in ½0; 1, reﬂecting the fact that some sensorial information is weakly accessible to consciousness, i.e. repressed. Mechanisms of memory storage and retrieval by the BM function as mentioned in Section 2. Once the network is initialized, we ﬁnd the stored patterns by presenting many random patterns to the BM, with an annealing schedule a that
allows stabilizing onto the many local minima of the network energy function. These initially stored patterns, associated as they are to two weakly linked subnetworks, represent neurotic memory states. To simulate working-through, we stimulate the net by means of a state change in a randomly chosen ni belonging to the ‘‘unconscious’’ (sensorial) set of a neurotic pattern. This stimulus is then presented to the network and, if the BM retrieves a pattern with symbolic conﬁguration different from that of the neurotic pattern, we interpret this as a new conscious association, and enhance all weights from ni to the changed nodes in the symbolic set. Following Hebbian learning, increment values are proportional to the products of the states of the connected neurons and the learning parameter b. New knowledge is learned only when the stimulus from the analyst is not similar to the neurotic memory trace. This procedure must be repeated for various reinforcement iterations in an adaptive learning process, and also each set of reinforcement iterations must be repeated for various initial temperature values.
4. Simulation illustrations and network behavior We illustrate the model with a simulation for a network with N ¼ 50 neurons, such that Nsens ¼ Nsymb ¼ 25 belong to each memory. In the brain, memory modules are not symmetric and neurons and synapses function differently in different brain regions. We concentrate on the interactions between and within the sensorial and symbolic modules to describe how the possibility of memory traces to become conscious is sensitive to the connection intensities, and how they can be reconﬁgured in a learning process to enhance or inhibit associativity. Although N is extremely small, initially, it seems to sufﬁce to illustrate the basic concepts and mechanisms at a semantic level. The short-range microscopic mechanisms are scalable and, since our algorithms are based on microscopic biological mechanisms, we expect that this level should be amenable to mapping to a biological substratum. Some simulations of system conﬁgurations have taken a few days on a single processor, even for these small systems. We plan to parallelize these algorithms, in order to simulate signiﬁcantly larger systems. In our simulations, s ¼ 0:58, b ¼ 0:3, Z ¼ 0:1 and the other parameters for annealing in the BM are attributed the same values we have used in . To conﬁgure neurotic traces having weak connections to symbolic processing brain areas, synapses connecting different memory modules are multiplied by k ¼ 0:5, deﬁning the initially stored patterns as neurotic. For smaller values of k the network has learning difﬁculties, suggesting a difﬁculty in associating unconscious traces among themselves and with symbolic memory. In Fig. 1(a), we show one topology generated after executing only the clustering algorithm and, in Fig. 1(b), the corresponding topology after long-range synaptic generation. The ﬁgures illustrate the clustered, hierarchical self-organization of the network. For this initial topology, 42% of the patterns stored initially were still stored after working-through, which shows that the network adapts with the reconﬁguration, freeing itself from some of the ‘‘neurotic’’ states. Figs. 2(a) and (b) give a more quantitative view. We generated 10 000 different neurotic topologies, for different values of N and spatial dimensions, from the same initial system parameter values speciﬁed above, and measured the average node degree (k) distributions. In Fig. 2(a) (in logarithmic scale) the discrete symbols represent the values found in our simulations. For larger values of N, Fig. 2(a) shows a power law distribution for large k, with exponent g % À4. This power law behavior is characteristic of many biological
Author's personal copy
ARTICLE IN PRESS
3370 R.S. Wedemann et al. / Neurocomputing 71 (2008) 3367–3371
7 6 5 4
positions sensorial memory positions declarative memory clustered synapses in memory
7 6 5 4
positions sensorial memory positions declarative memory synapses connecting memories
3 2 1 0 -0.5 0 0.5 1 x 1.5 2 2.5 3
3 2 1 0 -0.5 0 0.5 1 x 1.5 2 2.5 3
Fig. 1. (a) On the left, network topology after clustering for N ¼ 50. (b) On the right, network topology with long-range synapses.
10000 Average number of nodes with k links 1000 100 10 1 0.1 0.01 0.001 0.0001 1e-05 1 10
Average clustering coefficient
10**7/x**4.1 N = 32 N = 50 N = 100 N = 282 N = 1000 N = 2000 N = 4000
0.1 100 0 1000 2000 3000 4000 5000 Number of links (k) Total number of neurons (N)
Fig. 2. (a) On the left, node degree distributions in logarithmic scale, for various values of N. (b) On the right, average clustering coefﬁcient for different N values.
systems and indicates scale independence. It is known that random graphs follow the Poisson distribution of node degrees lk expðÀlÞ=k! . The deviations from Poisson in Fig. 2(a) for higher k values may be attributed to the competitive biological mechanisms, and conﬁrm that the resulting networks have a structure which is not random and may be scale free [16,19]. The average clustering coefﬁcient C [16,19] for 10 000 networks of 50 neurons is 0.38. We compare C generated by our clustering algorithm to that of a real network, where we have considered the neural network of the well-studied worm C. elegans, which is composed of 282 neurons. This worm has measured C ¼ 0:28 . We obtained an average C ¼ 0:24 within our model over 10 000 networks, also of 282 neurons. These two values are surprisingly similar, considering that our algorithm is based on very general microscopic biological mechanisms. This worm’s network would present C ¼ 0:049, if it were a random graph , and the higher measured value indicates the existence of network structure. In Fig. 2(b) crosses depict the calculated average C values, for the initial network conﬁgurations, as a function of N. As it is well ´ ¨ known, C is negligible in the case of an Erdos–Renyi random network. We also notice that the dependence for larger N is much weaker than that for a network in which linkage is performed
through the preferential attachment criterium, for which C / 1=N . To illustrate this point, we show as a full line in the graph the function 1=N 0:23 . We notice that, even if it follows the main trend of C obtained in our simulations, this function overpredicts the drop for large N. The clustering algorithm presented in this work thus appears to lead to networks, for which the biological mechanisms establish some different topological properties than the traditional algorithms, studied in complex network theory.
5. Conclusions We have further developed a previous memory model  to include microscopic biological neuronal mechanisms, and veriﬁed that the memory complex network self-organizes into a hierarchical clustered structure. Two hierarchically structured modules corresponding to sensorial and symbolic memories interact, producing sensorial and symbolic activity, representing unconscious and conscious mental processes. This memory structure and functioning, with an adaptive learning process, is used to explain a possible mechanism for neurotic behavior and psychoanalytical working-through.
Author's personal copy
ARTICLE IN PRESS
R.S. Wedemann et al. / Neurocomputing 71 (2008) 3367–3371 3371
The model is in agreement with psychoanalytic experience that working-through is a slow process, where the individual slowly elaborates knowledge by re-associating weakly connected memory traces and new experiences. It thus illustrates a phenomenological view of mind, with intentionality . Language is an important process, and is represented by the functioning of symbolic memory. Repetition is also important, and is illustrated in the model by the repetitive adaptive reinforcement learning, in the simulation of working-through. From the simulations, we saw that the working-through when understood as a network reconﬁguration process, partially frees the analysand from the original neurotic states and respective acts. The new network topology which results from this reconﬁguration process will stabilize onto new energy minima, associated with new conscious or unconscious acts. It is the purpose of the analysand and the analyst that these new minima are reached and generate acts which are more pleasant and comfortable to the analysand. The analysand is rewriting his life history, i.e. his past, through a process of new signiﬁcations, and his present and future by creating new possibilities . This new story is now embedded, i.e. written, in his biological neural network. We are proceeding in further model reﬁnement and analysis. It is necessary to verify more deeply the dependence of model behavior on parameters such as T, s and k. These parameters represent, at least partially, the effects of neurosubstances such as neural growth factors and neuromodulators. Although we do not yet have experimental indications of their values, tuning their relative values is fundamental for model stability and functioning. This can give some insight to basic mechanisms in real neural networks and the emergence of behavior. We are also interested in trying to map language structure and processing into network topology and dynamics, although we are not yet sure whether this is possible. Although biologically plausible, in accordance with many aspects described by psychoanalytic clinical experience, and experimentally based on simulations, the model is very schematic. We do not sustain or prove that this is the actual mechanism that occurs in the human brain. It nevertheless seems to be a good metaphorical view of facets of mental phenomena, for which we seek a neuronal substratum, and suggests directions of search. Our investigations strongly indicate the importance of the connection of symbolic processing, meaning and language for consciousness. References
 I. Aleksander, H. Morton, Phenomenology and digital neural architectures, Neural Networks 20 (2007) 932–937.  B. Baars, In the theatre of consciousness: global workspace theory, a rigorous scientiﬁc theory of consciousness, J. Consciousness Stud. 4 (4) (1997) 292–309.  F. Brentano, Psychology from an Empirical Standpoint, Routledge, London, 1995 (German ed. in 1874).  L.A.V. Carvalho, D.Q. Mendes, R.S. Wedemann, Creativity and delusions: the dopaminergic modulation of cortical maps, in: Proceedings of ICCS 2003, Lecture Notes in Computer Science, vol. 2657, Springer, Berlin, 2003, pp. 511–520.  G.M. Edelman, Wider than the Sky, a Revolutionary View of Consciousness, Penguin Books, London, 2005.  J. Forrester, The Seductions of Psychoanalysis, Freud, Lacan and Derrida, Cambridge University Press, UK, 1990.  S. Freud, The Unconscious, Standard Edition, vol. XIV, The Hogarth Press, London, 1957 (First German ed. in 1915).  S. Freud, Project for a Scientiﬁc Psychology, Standard Edition, vol. I, The Hogarth Press, London, 1966 (First German ed. 1950).  S. Freud, Introductory Lectures on Psycho-Analysis, Standard Edition, W. W. Norton & Company, New York, London, 1966 (First German ed. 1917).  S. Freud, Beyond the Pleasure Principle, Standard Edition, vol. XI, The Hogarth Press, London, 1974 (First German ed. 1920).  A. Fronczak, J.A. Holyst, M. Jedynak, J. Sienkiewicz, Higher order clustering ´ coefﬁcients in Barabasi-Albert networks, Physica A 316 (1–4) (2002) 688–694.  W.F. Ganong, Review of Medical Physiology, McGraw-Hill, USA, 2003.
 J.A. Hertz, A. Krogh, R.G. Palmer (Eds.), Introduction to the Theory of Neural Computation, Lecture Notes, vol. I, Santa Fe Institute, Studies in the Science of Complexity, Perseus Books, Cambridge, MA, 1991.  E.R. Kandel, J.H. Schwartz, T.M. Jessel (Eds.), Principles of Neural Science, McGraw-Hill, USA, 2000.  M. Kinsbourne, Taking the project seriously: the unconscious in neuroscience perspective, Ann. NY Acad. Sci. 843 (1) (1998) 111–115.  M.E.J. Newman, Random graphs as models of networks, in: S. Bornholdt, et al. (Eds.), Handbook of Graphs and Networks, from the Genome to the Internet, Wiley-VCH, USA, 2003, pp. 35–68.  M. Spitzer, A cognitive neuroscience view of schizophrenic thought disorder, Schizophr. Bull. 23 (1) (1997) 29–50.  J.G. Taylor, Mind and consciousness: Towards a ﬁnal answer?, Phys. Life Rev. 2 (2005) 1–45.  D.J. Watts, S.H. Strogatz, Collective dynamics of ‘small world’ networks, Nature 393 (1998) 440–442.  R.S. Wedemann, L.A.V. Carvalho, R. Donangelo, A complex neural network model for memory functioning in psychopathology, in: Proceedings of ICANN 2006, Lecture Notes in Computer Science, vol. 4131, Springer, Berlin, 2006, pp. 543–552.  R.S. Wedemann, L.A.V. Carvalho, R. Donangelo, Complex Networks in Psychological Models, Progress of Theoretical Physics Supplement, vol. 162, Yukawa Institute for Theoretical Physics, The Physical Society of Japan, Kyoto, 2006, pp. 121–130.  R.S. Wedemann, R. Donangelo, L.A.V. Carvalho, I.H. Martins, Memory functioning in psychopathology, in: Proceedings of ICCS 2002, Lecture Notes in Computer Science, vol. 2329, Springer, Berlin, 2002, pp. 236–245. Roseli S. Wedemann received her undergraduate degree in Physics in 1985, her Master of Science degree in Physics in 1988, and her Ph.D. in Systems Engineering and Computer Science in 1995, at the Federal University of Rio de Janeiro (UFRJ). She has been visiting professor at the Institute of Physics at UFRJ in 1996 and is currently professor at the Computer Science Department of the Institute of Mathematics of the State University of Rio de Janeiro (UERJ). She has been active in the research group in Computational ´ Neuroscience with Prof. Luıs Alfredo Vidal de Carvalho since 2000. She is referee of the Brazilian National Research Council since 2001 and of the Rio de Janeiro State Research Foundation since 2002. Her main research interests include Computational Neuroscience, Neural Networks, Artiﬁcial Intelligence, Distributed Algorithms, Computational Science, Physics of Complex Systems and High Performance Computing. ´ Luıs Alfredo Vidal de Carvalho received his undergraduate degree in Mechanical Engineering in 1982, his Master of Science degree in Optimization in 1984, and his Ph.D. in Systems Engineering and Computer Science in 1989, at the Federal University of Rio de Janeiro (UFRJ). He has acted as Chairman of the Graduate Program of Systems Engineering and Computer Science at UFRJ in 1995–1996 and 2000–2001, where he is full professor. He is currently Chairman of the Interdisciplinary Graduate Program of History of Science and Epistemology at UFRJ, since 2000. He has received the Latin America Biological Psychiatry Award in 1998, awarded by The World Biological Psychiatry Society, The Latin America Biological Psychiatry Society and Lilly - Neuroscience Pharmaceutical Laboratory, and also the IEEE - New Educational Technologies Award in 2000. He has lead the main research group in Computational Neuroscience in UFRJ since 1993. He is member of the New York Academy of Sciences since 1995, and referee of the Brazilian National Research Council since 1992. His main research interests include Computational Neuroscience, Neural Networks, Artiﬁcial Intelligence, Data Mining and History of Neuroscience and Epistemology. Raul J. Donangelo is born in Montevideo, Uruguay in 1947. He graduated as an industrial engineer at the ´ Universidad de La Republica, Montevideo, Uruguay, in 1971. He received a Ph.D. in Physics from the University of California at Berkeley in 1977. He has worked at the ´ Instituto de Fısica da Universidade Federal do Rio de Janeiro since 1978, and is presently working as a full professor. His main areas of research are theoretical nuclear physics and physics of complex systems.