Functional Implications of Temporal Structure in Primate Cortical Information Processing

download Functional Implications of Temporal Structure in Primate Cortical Information Processing

of 15

Transcript of Functional Implications of Temporal Structure in Primate Cortical Information Processing

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    1/15

    evant circuitry can be quickly adapted to the currentsignal processing requirements. I will show that dy-namic changes in synchronization patterns of neuronalpopulations are a good candidate for a mechanism thatdelineates and structures neuronal assemblies which arethe basis for distributed representation and processingunderlying cognitive processes.

    Processing and representation of informationin the mammalian cortex

    The mechanisms of neuronal processing clearly dependon the kind of coding used to represent information inthe nervous system. After strong emphasis on the im-

    REVIEW

    Functional implications of temporal structure in primate corticalinformation processing**

    Andreas K. Kreiter*

    Institut fr Hirnforschung, Universitt Bremen, Germany

    Summary

    Processing of information in the cerebral cortex of primates is characterized by distributed representations and processing in neuronalassemblies rather than by detector neurons, cardinal cells or command neurons. Responses of individual neurons in sensory corticalareas contain limited and ambiguous information on common features of the natural environment which is disambiguated by compari-son with the responses of other, related neurons. Distributed representations are also capable to represent the enormous complexity andvariability of the natural environment by the large number of possible combinations of neurons that can engage in the representation ofa stimulus or other content. A critical problem of distributed representation and processing is the superposition of several assemblies ac-tivated at the same time since interpretation and processing of a population code requires that the responses related to a single represen-tation can be identified and distinguished from other, related activity. A possible mechanism which tags related responses is the syn-chronization of neuronal responses of the same assembly with a precision in the millisecond range. This mechanism also supports the

    separate processing of distributed activity and dynamic assembly formation. Experimental evidence from electrophysiological investi-gations of non-human primates and human subjects shows that synchronous activity can be found in visual, auditory and motor areas ofthe cortex. Simultaneous recordings of neurons in the visual cortex indicate that individual neurons synchronize their activity with eachother, if they respond to the same stimulus but not if they are part of different assemblies representing different contents. Furthermore,evidence for synchronous activity related to perception, expectation, memory, and attention has been observed.

    Key words: synchronization, oscillation, gamma band response, macaque monkey, neuronal assembly

    0944-2006/01/104/0304-241 $ 15.00/0

    Introduction

    The way information is processed in the mammalianbrain is still one of the central, essentially unresolvedissues of science. While electrophysiological and neu-roanatomical investigations and more recently func-tional magnetic resonance imaging (fMRI) provided anincreasingly detailed picture of the functions particularregions of the brain are dealing with, there is only lim-ited understanding of the mechanisms by which theneuronal networks process information, even for simplecognitive tasks. In the following, I will argue that un-derstanding information processing in the neuronal cir-cuitry depends to a large extent on understanding howsignals are channeled through the brain and how the rel-

    *Corresponding author: Prof. Dr. Andreas Kreiter, Institut fr Hirnforschung, Universitt Bremen, Fachbereich 2, Postfach 33 04 40,

    D-28334 Bremen, Germnay; phone: ++49-421-2189093; fax:++49- 421-2189004, e-mail: [email protected]**Presented at the 94th Annual Meeting of the Deutsche Zoologische Gesellschaft in Osnabrck, June 48, 2001

    Zoology 104 (2001): 241255 by Urban & Fischer Verlaghttp://www.urbanfischer.de/journals/zoology

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    2/15

    portance of single units as a representation of stimuli inearlier research (Barlow, 1972), there is now over-whelming evidence in favor of distributed modes ofprocessing (Hebb, 1949; Edelman and Mountcastle,1978; Braitenberg, 1978; Grossberg, 1980; Hopfield,1982; Abeles, 1982; Aertsen et al., 1986; Zipser andAndersen, 1988; Gerstein et al., 1989; Georgopoulos,1990; Palm, 1990; Singer, 1990a, b; Abeles, 1991;Rolls, 1992; Young and Yamane, 1992). In a distributedsystem information does not converge on a single detec-tor neuron which would provide by its activity the entirerepresentation of the object for which it is the detector.Distributed processing modes are instead characterizedby the distribution of the signals describing a certain ob- ject (or other content) over many different neurons at

    different processing stages. Such contents may be forexample the picture of an object, the memory of it or amotor plan for grasping it. Individual neurons representby their responses the limited number of elementary fea-tures that constitute the usually much more complexcontents to be processed. The possibility to recruit themin an almost unlimited number of combinations intoneuronal assemblies allows the neuronal circuitry tohandle the enormous variety and complexity of the con-tent it needs to represent and process with a limitednumber of neurons (Freiwald et al., 2001). While mod-els based on distributed processing differ in their indi-vidual characteristics and capabilities, there are some

    common properties which make them a suitable modelfor neuronal processing. (1) The combinatorial nature ofa distributed code results in a coding capacity of a set ofneurons that is much higher than for the same number ofdetector neurons. (2) Distributed coding schemes canpreserve the individual features of an object which needto be lost by detector neurons due to their generalizingproperties (Kreiter and Singer, 1996a). (3) The associa-tive nature of many distributed processing schemes pro-vide for associative pattern completion (and error cor-rection), a characteristic property of mammalian percep-tion. (4) Fast processing based on rate coding dependson a distributed representation since reading a rate with

    sufficient resolution requires measuring intervals wellbeyond the time spans available. (5) Distributed pro-cessing architectures provide simple and biologicallyplausible mechanisms to generate new functional struc-tures by learning. Gradual changes of synaptic efficacyin a network of cooperating neurons are much less de-manding than the de novo generation of the entirewiring scheme for a fresh, so far unused neuron whichshould become a new detector neuron.Experimental investigations suggest indeed, that themammalian brain is best described as a distributed pro-cessing system. A paradigmatic example for a highly

    distributed system is the primate visual system. The vi-sual cortex ofMacaca mulatta, one of the most thor-

    oughly investigated sensory systems, consists of morethan 30 areas (Felleman and Van Essen, 1991). Theydiffer by the response properties of their neurons and bythe pattern of their rich interconnections with corticalas well as subcortical structures. A single visual stimu-lus results in activation of a large amount of neurons inmany cortical areas processing different aspects of thisstimulus. Individual neurons respond selectively tostimulus properties like position, orientation, spectralcomposition, binocular disparity, spatial and temporalfrequency, and direction of motion (Hubel and Wiesel,1959, 1962; Zeki, 1975; Orban, 1984; Desimone et al.,1985; Henry, 1985; Maunsell and Newsome, 1987;Livingstone and Hubel, 1988). Selective responseshave been observed for more complex stimulus proper-

    ties as well. This includes for example the componentsof optical flow fields (Saito et al., 1986), polar and hy-perbolic gratings (Gallant et al., 1993), geometric ar-rangements and patterns (Sakai and Miyashita, 1991;Tanaka, 1993, 1996; Janssen et al., 1999, 2000; Missalet al., 1999), simple objects (Logothetis et al., 1995),and even faces (Gross et al., 1972; Bruce et al., 1981;Perrett et al., 1985; Desimone, 1991; Rolls, 1992).While in particular the latter example has been taken asevidence for the concept of detector neurons, detailedinvestigations of their properties did not provide muchsupport for detector properties of these neurons. Facecells do not detect an individual's face (Desimone,

    1991) but are tuned to certain properties of faces likethe distance between facial features (Yamane et al.,1988) or the direction of gaze (Perrett et al., 1985). Ingeneral, neurons in the visual cortex respond like selec-tive filters, tuned for several, eventually complex stim-

    242

    A. K. Kreiter

    Zoology 104 (2001) 34

    Fig. 1. Ambiguity of information contained in the firing rate ofindividual neurons: The firing rate (r) of neurons in the visual cor-tex depends on multiple stimulus dimensions (D1 Dn). Theschematic example shows a 2-dimensional tuning curve for twodifferent stimulus properties (like orientation and spatial fre-quency) for which the neuron is tuned. Asuming a response rateof 23 Hz all the combinations of stimulus conditions depicted by

    the ring shaped trajectory in the stimulus space would result inthis particular response.

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    3/15

    ulus dimensions while ignoring others. This results inan ambiguous relation between firing rate and stimulisince the same firing rate can be evoked by many dif-ferent combinations of stimulus properties (Fig. 1).Disambiguation of the information contained in the fir-ing rate of individual neurons and precise representa-tion of certain stimulus properties can only achieved byconsidering the combination of responses in a popula-tion of neurons engaged in the processing of a stimulus.Taken together, experimental results as well as theoreti-cal considerations support the notion of distributed pro-cessing in the mammalian brain.

    The problem: dynamic structuring of a neuronal

    multi-purpose processing architectureOne of the central problems of distributed processingarchitectures is to organize the cooperation and signalexchange between the individual elements of the net-work. Not only in case of the mammalian cortex thenumber of different neuronal circuits that may beneeded is far too extensive to implement all of them inparallel independent sets of neurons with fixed internalconnectivity. Depending on the processing require-ments posed by external stimuli and internal state therequired networks rather need to be configured dynami-cally as necessary. In fast succession changing but

    overlapping sets of neurons have to contribute in differ-ent effective wiring schemes to changing networks withdifferent signal processing properties. Thus the cir-cuitry just needed has to be formed out of the given setof neurons and the rich network of their anatomicalconnections which may be considered as the limitingsuperset of all possible wiring schemes that may existin a given brain.Since anatomy is on a psychophysical time scale essen-tially fixed and can not change, there are several prob-lems that must be solved by a system based on dynamicdistributed processing. First, the computational proper-ties of a neuronal network depend to a large extent on the

    pattern of synaptic weights describing the connectivitymatrix (Braitenberg, 1978; Palm, 1982; McClelland andRumelhart, 1986; Rumelhart and McClelland, 1986).Therefore, the effective strength of a connection be-tween two neurons will depend on the type of ensembleto which they just contribute. Under different conditionsrequiring different ensembles with different connectivitymatrix this strength may need to be changed. Second, adynamically established ensemble of neurons which ful-fills a certain function will in general need only a subsetof the anatomical connections which terminate on itsneurons. Thus many signals which arrive in the network

    are not related to the coherent set of signals just pro-cessed. These unrelated signals act as noise disturbing

    the processing of the coherent signals and should there-fore be suppressed. Third, functionally distinct ensem-bles processing,e.g.,different stimuli need not only toavoid mutual influences but have in addition to providetheir efferent signals containing the results of their pro-cessing in a format that can be distinguished and selec-tively processed by other ensembles receiving inputfrom both of them. Thus, the central problem of dynami-cally structuring networks of different functions is theflexible routing and association of signals in an essen-tially fixed anatomy to allow neurons to be combinedinto temporary networks of variable composition.Early models of distributed processing are based onDonald Hebbs (1949) model of the cell assembly. Ac-cording to this model neurons that are often co-acti-

    vated because they get activated by the same stimulusare thought to strengthen their mutual synaptic connec-tions. This results in a network with properties of an as-sociative memory since partial activation of an assem-bly will lead to its full recruitment by the stronger mu-tual connectivity between neurons belonging to thesame assembly (Braitenberg, 1978; Palm, 1982). Themutual enhancement of activity between neurons be-longing to the same assembly may also serve to en-hance the visibility of the representation of an objectdue to the contrast between the stronger activity of neu-rons within an assembly and the weak activity of neu-rons not being part of an active assembly. This will in

    turn contribute to a reduction of disturbing signals fromsuch unrelated neurons and an increase of the signal-to-noise ratio for the processing of signals within the as-sembly. This model uses the available number of neu-rons economically since a given neuron may be part ofmany different assemblies.Even though the Hebbian assembly shows several ofthe above mentioned properties expected from a viablemodel of distributed processing, it suffers from twoshortcomings. The first is the fixed effective connectiv-ity which can not be adapted quickly because it is deter-mined by synaptic strength. The synaptic connectionsmay be modified by learning, but not on the fast time-

    scale of changing stimulus constellations or otherwisealtered cognitive states. This limits the flexibility withwhich different processing capabilities can be achievedfrom a given set of neurons. Second, the simultaneousprocessing of two different stimuli will in general notbe possible, because the Hebbian assembly fails to keeprepresentations of different stimuli separable, leading tothe so-called superposition catastrophe (von der Mals-burg, 1981, 1985). The interpretation of a distributedrepresentation requires the identification of the set ofneurons which contributes to this representation and todistinguish them from other neurons contributing to the

    representation of other stimuli. The neurons of a Heb-bian assembly may be identified as members of an acti-

    243

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    4/15

    vated assembly by their enhanced activity, but they cannot be distinguished by this criterion from the neuronsbeing part of another active assembly which representsa second stimulus. Therefore, another processing stagereceiving input from neurons of both assemblies wouldbe confronted with the superposition of two patternswhich is not interpretable (Fig. 2).The superposition problem can not in general be re-solved by the assumption of specific anatomical con-nections between different processing stages of a stimu-lus because of the massive convergence and divergenceof neuronal connections in the cortex. It is well knownthat each processing stage receives synaptic connectionsfrom a wide range of different neurons placed in its im-mediate neighborhood, the same cortical area or differ-

    ent areas. These neurons differ considerably in their re-sponse properties and are therefore likely to be part ofdifferent neuronal assemblies if complex, natural stimu-lus constellations have to be processed. A solution byexclusive anatomical connections from an assembly to acertain target structure is also excluded by the overlap ofdifferent assemblies that get activated by different stim-ulus constellations. A set of neurons which constitute anassembly in a given stimulus constellation may becomepart of several different assemblies with the next stimu-lus constellation to be processed only a tenth of a secondlater. By this change of relations between the neurons adifferent set of axonal connections would be required, if

    afferent connections would have to be organized in amanner depending on the composition of the set of as-semblies just activated.Taken together, the difficulties to interpret distributedcodes in the presence of multiple stimulus representa-tions and to configure dynamically the effective neu-ronal circuitry indicate the necessity of mechanism thatlabels neuronal responses to distinguish their origina-tion from different stimulus representations. Such alabel should facilitate differential processing of re-sponses from different assemblies.

    Labeling neuronal responses a role fortemporal structure of neuronal activity

    The unavoidable lack of specificity of fixed andwidespread anatomical connections originating from agiven assembly with respect to the dynamically chang-ing composition of other neuronal assemblies requiresthe responses themselves to carry the necessary infor-mation to identify their association with each other. Afirst possibility for such a label would be the firing rate.Neurons of different assemblies could express their re-lationships by firing rates that are identical for the neu-rons of the same assembly but differ for different as-semblies. Such a mechanism faces several problems:First, the limited range of possible firing rates and the

    244

    A. K. Kreiter

    Zoology 104 (2001) 34

    Fig. 2. The superposition problem of neuronal assemblies and its possible solution by synchronization: Different objects of a visualscene (A, B) will activate different populations of neurons which represent their properties. If the neurons of these populations providelargely overlapping anatomical connections to overlapping sets of target neurons, there is no possibility for the recipient to distinguishresponses related to different populations (C). Thus, an individually distributed representation could not be accessed, if several popula-tions are active simultaneously. However, the superposition of the responses elicited in different populations can be resolved, if neurons

    related to the same population discharge synchronously and avoid synchronization with other populations (D). Different populationsmay therefore be active simultaneously without confounding their stimulus descriptions.

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    5/15

    small resolution of rate estimates in short time intervalssuggests that only a very small number of assembliescan be distinguished by rate differences. Second, it isgenerally accepted that the firing rate of an individualneuron contains information about the properties of theactivating stimulus. This information would be lost, ifthe neurons of the same assembly would all fire withthe same rate. Third, experimental findings do not sup-port the assumption that neurons respond only in dis-crete frequency steps but show that they are continu-ously dependent on various stimulus properties like ori-entation, direction or spectral composition. For thesereasons a simple rate-based mechanism to distinguishresponses from different assemblies has to be rejected.Another property of spike trains that could be used to

    distinguish neuronal responses are the seemingly ran-dom fluctuations of their precise temporal structure.Contrary to spike rate there is little evidence that thesefluctuations serve to encode stimulus properties. Thishas led to the hypothesis that fine temporal structurecould serve to express the relation of neurons that be-long to the same or to different activated assemblies(Milner, 1974; Grossberg, 1980; Abeles, 1982; von derMalsburg, 1985, 1986; von der Malsburg and Schnei-der, 1986; von der Malsburg and Singer, 1988; Gersteinet al., 1989; Singer, 1990a, 1999; Abeles, 1991; Singerand Gray, 1995; Kreiter and Singer, 1996a; Freiwald etal., 2001). Neurons that engage in an episode of syn-

    chronous activity would thereby identify their dis-charges as part of a population-coded signal producedby the distributed processing of an assembly. Neuronsof a second assembly would also synchronize the tem-poral structure of their activity within the range of mil-liseconds but would avoid synchronization with thefirst and any other assembly (Fig. 2D). Precise synchro-nization would therefore become a tag for signalsevoked in the same assembly. Such a code does notneed to compromise the rate code containing stimulusspecific information. Synchronization or de-synchro-

    nization only require to shift individual spikes by a fewmilliseconds backward or forward in time but do notneed a change of their average probability of occur-rence which determines the rate.The primary effect of such differential synchronizationis that spikes which are part of the same activity patternthat represents a certain stimulus evoke simultaneouspost-synaptic potentials. This enhances the common ef-fect of the respective assembly on the post-synapticneuron by spatial summation at the dendritic tree. Fur-thermore, the synchronous arrival of the synaptic po-tentials facilitates the interactions and hence the com-mon and cooperative processing of the signals originat-ing from the same stimulus representation. The concen-tration in time of the occurrence of related synaptic po-

    tentials also reduces the probability of interaction withunrelated signals and increases the proportion of postsynaptic activity on a dendritic tree that is related to asingle stimulus. This separates different computationalprocesses in time that need to be processed concur-rently and enhances the signal-to-noise ratio for the in-dividual computation at the dendritic tree.A further advantage of neuronal assemblies defined bysynchronous activity is their ability to change their extentin a flexible, stimulus dependent manner. Simulationstudies demonstrated that synchronization can be used toselect responses into an assembly that represent the sameperceptual object and to segregate them from responses

    evoked by different objects or the background (Wang etal., 1990; Knig and Schillen, 1990; Grossberg andSomers, 1991; Horn and Usher, 1991; Sporns et al., 1991;Arndt et al., 1992; Neven and Aertsen, 1992; Schillenand Knig, 1994; Sompolinsky and Tsodyks, 1994; Ritzet al., 1994). Neuronal assemblies defined by syn-chronous activity could therefore serve to bind the repre-sentations of different features of an object into a coher-ent assembly that provides for the necessary exchangeand convergence of information which is a necessary pre-condition for the perception of a coherent object.

    245

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

    Fig. 3. Synchronized spike activ-ity of neurons recorded in area MTof a fixating macaque monkey: Thetwo examples were both recordedwith an electrode acquiring spikesfrom a few different, nearby neu-rons which respond to a movingbar stimulus. Spikes from differentneurons tend to occur in clusterseparated by silent intervals. Thisclustered appearance is reflected inthe auto-correlograms below by thebroad center peak and the initial

    troughs (modified from Kreiter andSinger, 1996a).

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    6/15

    Experimental evidence for the correlationhypothesis from the primate brain

    Synchronous states of neuronal activity

    The first prediction of the hypothesis described aboveis the synchronization of action potentials for neuronsthat respond to the same stimulus. Figure 3 shows ashort episode of a multi-unit recording from area MT ofan awake fixating monkey. The differently sized actionpotentials of different neurons engage for a short epochin a typical burst and pause pattern of activity. Withinsuch a multi-unit burst the spikes of different neuronsoccur within a few milli-seconds. Thus, a local group ofneurons recorded from a single micro-electrode can in-deed strongly synchronize their activity, if activated by

    the same stimulus. Such strongly synchronized activityin area MT is typically associated with a more or lessregular pattern of oscillatory activity with the fre-quency in the range of the gamma-band of the EEG(~3060 Hz) (Kreiter and Singer, 1992).In the local field potential, which is thought to reflectmainly the summation of synaptic currents and is there-fore a measure of local average activity, distinct spin-dles of activity in the same frequency range have beendescribed. They are phase-locked with the local spikeactivity if both signals are recorded simultaneously(Murthy and Fetz, 1992; Kreiter, 1992; Eckhorn et al.,1993; Livingstone, 1996). These observations suggestthat synchronization does not only occur pair-wise.Rather whole groups of neurons can synchronize theirfiring pattern in a manner that causes sequences of dis-charges synchronized over entire groups of neurons.Evidence for precise synchronization in the primatecortex has been described by several studies done inmacaque monkeys in the cortical areas V1, V2, V3, V4,MT, IT, the auditory, somatosensory and motor cortex(Aiple and Krger, 1988; Krger, 1990; Gochin et al.,1991; Bullier et al., 1992; Murthy and Fetz, 1992;Ahissar et al., 1992; Kreiter and Singer, 1992; Eckhornet al., 1993; Livingstone, 1996; Mller et al., 1996;deCharms and Merzenich, 1996; Freiwald et al., 1998;Nowak et al., 1999; Steinmetz et al., 2000). Syn-chronous activity has not only be observed locally butalso between neurons recorded from different sites inthe same area and different areas of the cortex (Frien etal., 1994; Munk et al., 2000).To be instrumental for tagging neuronal responses ofthe same assembly synchronization should be precisewithin the range of several milli-seconds. In accordancewith this requirement, the temporal precision of neu-ronal synchronization found in most of the studies is inthe range of a few milli-seconds. Much wider correla-tion peaks have been described, too (Gochin et al.,

    1991), but are not discussed here since they are likely toreflect different mechanisms.

    With respect to the dependence of neuronal synchro-nization for the response properties of neurons, the cor-relation hypothesis predicts a weak specificity of the in-teractions. Individual stimuli typically recruit neuronswith similar response properties but also activate neu-rons with quite different response properties, if theseneurons have a sufficiently wide tuning. If synchroniza-tion serves to define coherent groups of neurons thatrepresent a stimulus, it should occur between neuronswith similar and with dissimilar response properties.Experimental investigations have found synchroniza-tion preferentially between neurons with similar re-sponse properties like overlapping receptive fields,similar orientation or direction tuning but also betweenneurons with considerable differences between their

    properties. Recordings in area MT of awake macaquemonkeys revealed synchronization in cases of neuronswith differences in their preferred direction of motionof more than 90 (Kreiter and Singer, 1996b).Summarizing these results it can be stated that neuronalsynchronization occurs between the extended sets ofneurons that form neuronal representations and has asufficient temporal precision to serve as a label for dy-namic grouping of responses evoked from the sameneuronal assembly.

    Dynamic properties of neuronal synchronization

    One of the central features of distributed processingand representation is the flexible usage of the same neu-ron in a large number of different assemblies. Two neu-rons which are activated by the same stimulus and be-long therefore to the same assembly may be activated inthe next moment by two different stimuli. They willthen belong to two different assemblies which processdifferent stimuli in a different manner. This requires alabel of relatedness of neuronal discharges to changedynamically in dependence of the stimulus configura-tion. The prediction is that neurons that are synchro-nized, if activated by the same stimulus should not syn-chronize, if they are activated by different stimuli. This

    prediction contrasts with the earlier assumption thatcorrelated discharges reflect the anatomical propertiesof a network of neurons and are therefore largely inde-pendent on stimulus properties and of no particularfunctional relevance.Figure 4 shows the results of an experiment which hasbeen done to test this prediction in area MT of amacaque monkey which fixates a central fixation spot.The experiment requires simultaneous recordings fromneurons having overlapping receptive fields with dif-ferent but similar preferred directions of motion. Stimu-lation with a single bar moving in a direction between

    the two preferred directions of motion of the neurons atthe two recording sites (Fig. 4 A) activated neurons at

    246

    A. K. Kreiter

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    7/15

    247

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

    Fig. 4. Stimulus dependence of neuronal synchronization in area MT: (A, B) shows plots of the receptive fields and the stimulus con-figurations. The dot marked F corresponds to the fixation point. Arrows within the RF plots indicate the preferred direction of motionfor the neurons at the respective recording sites and arrows at stimulus bars the direction of motion. (A) depicts the single and (B) thedual bar configuration. Cross-correlogram and PSTHs obtained for both conditions are shown below (C, D). The thin vertical lines inthe PSTHs mark the window over which the cross-correlograms were computed. The scale bars correspond to 40 spikes/s. Note the pro-nounced synchronization in the single bar condition and the absence of synchronization in the dual bar condition. Scatter plots of nor-malized correlation values (NC, peak amplitude above offset divided by the offset) and firing rates obtained for the single bar condition(ordinate) against those obtained for the dual bar configuration (abscissa) are shown in (E) and (F), respectively. The dashed line indi-

    cates the region of equal values for both conditions. In all cases synchronization is considerably stronger for the single bar conditionwhile response rates are similar (modified from Kreiter and Singer, 1996a).

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    8/15

    248

    A. K. Kreiter

    Zoology 104 (2001) 34

    Fig. 5. Specificity of stimulus dependent synchronization: Comparison of synchronization in the single and crossing bar configuration.Conventions are the same as in Fig. 4. Plots of RFs, stimulus configuration, PSTHs and cross-correlogram are shown for the single- (A,C) and the crossing bar condition (B, D). In this case the normalized correlation (NC) was 56.5% for the single bar configuration, 58.8%for the crossing bar configuration and 4.4% for the dual bar configuration (data not shown). Scatter plots of normalized correlation (NC)and firing rates obtained for the crossing bar configuration (ordinate) against those obtained for the single bar condition (abscissa) areshown in (E) and (F), respectively. Note that the additional bar in the crossing bar configuration causes no major reduction of synchro-nization as compared to the reduction found for the dual bar configuration. Conventions as in Fig. 6 (modified from Kreiter and Singer,1996a).

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    9/15

    249

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

    Fig. 6. Dependence of synchronization on different directions of motion: Stimulus configurations are indicated in the plots above therespective cross-correlograms and PSTHs. Other conventions are the same as in Fig. 4. The single bar configuration (A, repeated in E)resulted in a strong correlation (NC = 73.9% in A and 52.1% in E) which essentially disappeared in the dual bar configuration (B, NC =3.8%). Changing orientation and direction of motion of the single bar by 15 resulted only in a minor change of correlation (C, NC =60.6%). In (D) the orientation and direction of motion of the single bar are changed further so that it equals the right bar of the dual barconfiguration shown in (B) and stimulates site 1 only poorly. Note that the NC value remains similar as in the original single bar condi-tion (NC = 56.8%). The scatter plot (F) of NC values obtained for presentation of one of the bars of the dual bar configuration (ordinate)

    versus those obtained for the single bar configuration indicates that in only a few cases the single bar condition resulted in a strongercorrelation (modified from Kreiter and Singer, 1996a).

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    10/15

    both sites. In this stimulus constellation the neurons atboth recording sites contribute to representation andprocessing of the same stimulus and are therefore con-sidered to be part of the same assembly. Stimulationwith two different bars, each of them moving in the pre-ferred direction of one of the receptive fields, activatesthe neurons at both recording sites too (Fig. 4 B). Ac-cording to the rules of gestaltpsychology for bindingvisual features into a coherent object the two bars lackall properties like common direction of motion orcollinearity that would bind them into a single coherentobject and are therefore perceived as two independentobjects. Therefore, the two stimuli are expected to berepresented by two different assemblies. In contrast tothe single-bar-condition the neurons at the two record-

    ing sites are not part of the same assembly for this dual-bar-condition but are expected to belong to the two dif-ferent assemblies, each one representing one of the twobars. In accordance with the prediction there is a clearcorrelation peak in the cross-correlogram obtained forthe single-bar-condition (Fig. 4 C), but no such peak isobserved in the cross-correlogram for the dual-bar-con-dition (Fig. 4 D). This result can not be attributed tochanges of the mean firing rate since firing rates do notvary systematically with the stimulus conditions in thisexperiment (Fig. 4 F).While this result is well in line with the predictions ofthe correlation hypothesis, two alternative interpreta-

    tions need to be ruled out. The first suggests that thelack of correlation in the dual-bar-condition does notreflect the relation of the neurons at the two recordingsites to different assemblies but a general reduction ofsynchronicity and temporal structure, possibly due tothe presence of two instead of one independent stimuli.To test this possibility the same configuration as in thesingle-bar-condition (Fig. 5 A) was extended with an-other bar crossing over both receptive fields (Fig. 5 B).This stimulus constellation is very similar to the dual-bar-condition, but the neurons at both recording sitesare activated by the same bar and not by two differentbars as in the dual-bar-condition. The correlation hy-

    pothesis predicts that despite the presence of two stim-uli the neurons at both sites should synchronize theiractivity with each other, since they represent the samestimulus. As shown in Figure 5 CF the experimentalresults are in line with this prediction. Therefore, thefirst alternative explanation has to be rejected.The second alternative explanation suggests that thesynchronization observed in the single-bar-conditiondepends essentially on the intermediate direction ofmotion of the bar. The neurons activated most stronglyby this stimulus direction are expected to provideshared inputs to the neurons at both recording sites and

    would thereby cause more synchronization of the neu-rons (Perkel et al., 1967; Moore et al., 1970; Gerstein

    and Perkel, 1972). To test the dependence of synchro-nization on direction of motion (Fig. 6) the single barcondition (Fig. 6 A) was presented with different direc-tions of motion between and including the direction ofmotion of the bars used in the dual bar condition (Fig. 6C, D). The results showed that even in the extreme caseof a bar moving in the same direction as one of the barsin the dual-bar-condition the neurons at both sitesstayed synchronized, if they were both activated by thisindividual bar. This result shows clearly that neither un-specific reductions of temporal coordination in thepresence of multiple objects nor specific stimulus prop-erties determine the pattern of synchronization. It is theactual membership of a neuron in an assembly thatcauses the neuron to synchronize its pattern of dis-

    charge with that of the other neurons in the assembly,but not with those which belong at the same time to dif-ferent assemblies.Evidence for state or stimulus dependent changes intemporal interactions of neurons has also been de-scribed in other parts of the monkeys cortex. Freiwaldet al. (1998) found that in the inferotemporal cortexsynchronization and oscillation strength can undergorapid changes in response to a stimulus, even if nochanges of firing rate were observed. Within the audi-tory cortex deCharms and Merzenich (1996) found anincrease in synchronization for the whole duration of asound stimulus between neurons that had only phasic

    responses to stimulus onset and offset on top of sponta-neous activity. Directional information for a reachingtask has been found to be represented by the synchro-nization between neurons in the motor cortex aroundthe time of motion onset (Hatsopoulos et al., 1998).States of attentive expectation have been found to in-crease synchronous activity in the macaques visual cor-tex areas MT and MST (Cardoso de Oliveira et al.,1997) and in the motor cortex (Murthy and Fetz, 1992;Sanes and Donoghue, 1993; Vaadia et al., 1995; Riehleet al., 1997; Grammont and Riehle, 1999) as well as be-tween different visual and motor areas (Munk et al.,2000). All these studies suggest that different cognitive

    processes are associated with the activation of neuronalassemblies characterized by synchronous discharge oftheir constituting neurons.

    Perception and synchronous high-frequency oscillatory-activity in the human brain

    If neuronal assemblies are characterized by syn-chronous activity in the gamma-band, it may be possi-ble to observe this activity in the EEG of human sub- jects. Even though this activity has been difficult tomeasure because of its small amplitude, it has now been

    observed by several laboratories. In accordance withthe correlation hypothesis and the results of animal

    250

    A. K. Kreiter

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    11/15

    studies enhanced gamma-band activity was found in re-sponse to coherently moving stimuli as compared to in-coherent stimuli (Lutzenberger et al., 1995; Mller etal., 1996). This reduction of gamma-band activity in theEEG for multiple independent stimuli does not neces-sarily indicate a lack of synchronous activity in neu-ronal assemblies. The results of single- and multi-unitrecordings in the macaques visual cortex describedabove show, that multiple, independently moving, andeven overlapping bar stimuli also result in synchro-nized activity within, but not between the respectiveneuronal assemblies. The most likely reason for the re-duced gamma-band activity observed for such condi-tions is therefore the averaging of EEG recordings overmultiple desynchronized assemblies. Due to the lack of

    synchrony between the individual assemblies the elec-trical signals of their neurons tend to cancel each otherwhereas they would add up when the neurons were partof a single, synchronized assembly. Therefore, differ-ences of gamma-band activity between similar stimulusconditions are usually interpreted as an indication for adifference in synchronicity of neurons within an ex-tended assembly or the difference between states withneurons being organized into one extended assemblyversus several smaller assemblies.Unfortunately, this interpretation of results of EEGstudies is not without problems. While enhancedgamma-band responses may well reflect increased

    synchronization of oscillatory activity within a givenset of cells, it could also result from an increase of thenumber of activated cells or an increase in their activ-ity without increased temporal synchrony. Dependingon the methods to evaluate synchronization betweendifferent EEG electrodes similar problems may arise.Some early measures of synchronization confoundedphase and amplitude information so that changes inoscillation amplitudes resulted in changed measures ofsynchrony. A more critical problem is that changes insynchrony between EEG electrodes may be caused bychanges in the neuronal populations that contribute tothese signals. Thus, a sudden increase in synchrony

    between electrode 1 and 2 does not necessarily reflecta change in the synchrony between the same twogroups of neurons. It may also result from the activa-tion or inactivation of another group of neurons, e.g.recorded by electrode 1 which always discharges insynchrony with neurons at electrode 2. The main argu-ments against such alternative explanations ofgamma-band responses and their correlation in EEGrecordings are the results of animal experiments whichdemonstrate that even well-isolated individual neu-rons can change dynamically their synchronization in-dependent of rate changes (Freiwald et al., 1995; Kre-

    iter and Singer, 1996b) and the consistency in the in-terpretation of results of many different studies which

    would not be expected, if alternative explanationswould often describe the dominant mechanism ofgamma-band responses.According to the correlation hypothesis the activationof such synchronously firing neuronal assemblies is theneuronal mechanism for the internal representation andprocessing of sensory information. Since object percep-tion critically depends on binding of the features of anobject, the correlation hypothesis predicts that objectperception should be associated with the synchroniza-tion of extended groups of neurons which represent bytheir activity elementarily features of an object. Thissuggests that object perception may be associated withenhanced gamma-band activity in the EEG. Well in linewith this prediction it has been shown that the percep-

    tion of the Kanizsa triangle (like a real triangle) resultsin a distinct phase of enhanced gamma-band activity, ifcompared with a similar display in which no triangle isperceived since the inducers of the Kanizsa triangle hadbeen rotated (Tallon et al., 1995; Tallon-Baudry et al.,1996). The perception of hidden figures in clutteredpictures often requires subjects to know what theycould see in such a picture. A comparison of thegamma-band responses between pictures that are mean-ingless to subjects and the identical pictures after sub-jects were trained to perceive a hidden dalmatian dogalso reveals enhanced gamma-band activity in the EEGto be associated with the perception of the dog (Tallon-

    Baudry et al., 1997). Similarly, the emergence of athree-dimensional percept while viewing a random-dotstereogram is associated with a transient gamma-bandpeak around 40 Hz, 300 to 500 ms before the subjectsreport the appearance of the perception (Revonsuo etal., 1997). The most direct evidence relating perceptionand synchronous activity in the human brain is pro-vided by estimations of the synchrony between EEGelectrodes covering the entire scalp (Rodriguez et al.,1999). In this study human subjects had to signal theperception of upright and upside-down mooney faces.For trials in which the subjects perceived the face a dis-tinct pattern of electrode pairs synchronized their

    gamma-band activity. On average the synchrony be-tween electrodes increased in such trials, while for tri-als leading to no perception of a face synchrony ratherdeclines. Taken together, these results suggest that per-ception is associated with a specific increase in syn-chronous gamma-band activity due to the activation ofextended assemblies binding responses related to theperceived object into a coherent representation.

    Synchronized activity related to memory and learning

    Internal representations are not only driven by external

    stimuli but may be activated during rehearsal and reten-tion processes for short-term memory. Therefore, Tal-

    251

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    12/15

    lon-Baudry et al. (1998, 1999) investigated the gamma-band responses in a delayed match to sample task. Theyfound sustained oscillatory activity in the gamma andbeta band within the delay period in which no stimuluswas present but the sample stimulus had to be kept inmemory. In contrast to representations in short-termmemory content contained in long-term memory is keptfor much longer periods of time. Donald Hebb (1949)suggested that this may be achieved by forming cell as-semblies from the neurons that are activated by the re-spective stimuli. In a classical conditioning paradigmthat associated an conditioned light stimulus (CS+) anda noxious unconditioned stimulus (UCS) to the skin,enhanced coherence between the gamma-band activityin the visual and the somatosensory cortex is observed

    during presentation of the CS+

    before the UCS as com-pared to CS trials (Miltner et al., 1999) after training.This suggests that that new assemblies formed by asso-ciative learning processes are also characterized bysynchronous activity in the gamma-band.

    Synchronized activity related to attention

    As pointed out before, synchronous activity does notonly tag a population with a label that can be used toidentify related sets of neurons, but also increases theeffectiveness of the synchronized responses at commontarget neurons. Since enhanced visibility of the neu-

    ronal activity generated for the representation of a sin-gle object has been suggested to be the neuronal corre-late for selective attention, changing the degree of syn-chronization within a neuronal assembly could be aneffective mechanism for selective attention (Niebur etal., 1993; Niebur and Koch, 1994). Indeed Gruber et al.(1999) found attentional modulation of the gamma-band response in a visual spatial attention task. How-ever, a recent study in area V4 of the macaque monkeyfound no significant changes in the correlation ofspikes or local field potentials but a small improvementof the temporal precision that link spikes of one record-ing site to the local field potential at a different record-

    ing site in the same area. Considerably stronger effectshave been found in the secondary somatosensory cortexof macaques which were trained to switch attention be-tween a visual task and a tactile discrimination task.The reason for these discrepancies is not yet clear. It ispossible that different modalities are differently af-fected by attention or that shifts of the task from onemodality to another one affect neuronal synchroniza-tion patterns more than spatial shifts of selective atten-tion within the same modality. Furthermore, it is possi-ble that the main effect of spatial selective attention isnot a change of synchronization between neurons of the

    same area but rather between neurons at different levelsof processing in different cortical areas. Selective cou-

    pling of restricted parts of the afferent population to tar-get neurons has been considered to be a key mechanismof spatial selective attention even though a differentlinking mechanism was proposed (Olshausen et al.,1993).

    Concluding remarks

    In this short review, I presented theoretical considera-tions and experimental results which suggest that infor-mation processing in the cerebral cortex is based onneuronal assemblies that are dynamically defined bythe synchronization of their constituting neurons. Thepresented evidence suggests that this mechanism servesto distinguish representations of multiple objects andorganizes information processing within the corticalcircuitry by setting the effective connectivity betweenindividual neurons. Therefore, this mechanism seemsto be a basic cortical function used for different cogni-tive processes like perception, memory and attention.Relevant experimental data for this hypothesis were notonly obtained in studies on human subjects or non-human primates but also in other vertebrate species likethe cat (for comprehensive review see: Singer andGray, 1995; Singer, 1999).

    Acknowledgements

    The author is grateful to Winrich Freiwald, Sunita Man-don, Catherine Tallon-Baudry, Katja Taylor and DetlefWegener for many stimulating discussions about sub-jects discussed in this paper and to Sabine Melchert forhelp in preparing the manuscript. This work was sup-ported by the Grant AZ I/76703 of the VolkswagenS-tiftung and the SFB 517 (Neurocognition) of theDeutsche Forschungsgemeinschaft.

    ReferencesAbeles, M. 1982. Local cortical circuits. Springer-Verlag, Berlin

    Heidelberg New York.Abeles, M. 1991. Corticonics. Cambridge University Press, Cam-

    bridge.Aertsen, A., G. Gerstein and P. Johannesma. 1986. From neuron

    to assembly: Neuronal organization and stimulus representa-tion. In: Brain Theory (G. Palm and A. Aertsen, eds.).Springer-Verlag, Berlin, pp. 724.

    Ahissar, M., E. Ahissar, H. Bergman and E. Vaadia. 1992. En-coding of sound-source location and movement: activity ofsingle neurons and interactions between adjacent neurons inthe monkey auditory cortex. J. Neurophysiol. 67: 203215.

    Aiple, F. and J. Krger. 1988. Neuronal synchrony in monkey

    striate cortex: interocular signal flow and dependency on spikerates. Exp. Brain Res. 73: 19.

    252

    A. K. Kreiter

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    13/15

    Arndt, M., P. Dicke, M. Erb, R. Eckhorn and H. J. Reitboeck.1992. Two-layered physiology-oriented neuronal networkmodels that combine dynamic feature linking via synchroniza-

    tion with a classical associative memory. In: Neural networkdynamics (J. G. Taylor, E. R. Caianello, R. M. J. Cotterill andJ. W. Clark, eds.). Springer, New York, pp. 140154.

    Barlow, H. B. 1972. Single units and cognition: A neurone doc-trine for perceptual psychology. Perception 1: 371394.

    Braitenberg, V. 1978. Cell assemblies in the cerebral cortex. In:Lecture Notes in Biomathematics 21, Theoretical Approachesin Complex Systems (R. Heim and G. Palm, eds.). Springer,Berlin, pp. 171188.

    Bruce, C. J., R. Desimone and C. G. Gross. 1981. Visual proper-ties of neurons in a polysensory area in superior temporal sul-cus of the macaque. J. Neurophysiol. 46: 369384.

    Bullier, J., M. H. J. Munk and L. G. Nowak. 1992. Synchroniza-tion of neuronal firing in areas V1 and V2 of the monkey. Soc.Neurosc. Abstr. 18: 11.

    Cardoso de Oliveira, S., A. Thiele and K. P. Hoffmann. 1997.Synchronization of neuronal activity during stimulus expecta-tion in a direction discrimination task. J. Neurosci. 17 (23):92489260.

    deCharms, R. C. and M. M. Merzenich. 1996. Primary corticalrepresentation of sounds by the coordination of action-poten-tial timing. Nature 381: 610613.

    Desimone, R. 1991. Face-selective cells in the temporal cortex ofmonkeys. J. Cogn. Neurosci. 3: 18.

    Desimone, R., S. J. Schein, J. Moran and L. G. Ungerleider. 1985.Contour, color and shape analysis beyond the striate cortex.Vision Res. 25: 441452.

    Eckhorn, R., A. Frien, R. Bauer, T. Woelbern and H. Kehr. 1993.High frequency (6090 Hz) oscillations in primary visual cor-tex of awake monkey. Neuro Report 4: 243246.

    Edelman, G. M. and V. B. Mountcastle. 1978. The Mindful Brain.MIT Press, Cambridge.

    Felleman, D. J. and D. C. Van Essen. 1991. Distributed hirarchi-cal processing in the primate cerebral cortex. Cereb. Cortex 1:147.

    Freiwald, W. A., A. K. Kreiter and W. Singer. 1995. Stimulus de-pendent intercolumnar synchronization of single unit re-sponses in cat area 17. Neuro Report 6: 23482352.

    Freiwald, W. A., A. K. Kreiter and W. Singer. 1998. Oscillatoryand synchronous activity states in the macaque inferotemporalcortex. Soc. Neurosci. Abstr. 24: 900.

    Freiwald, W. A., A. K. Kreiter and W. Singer. 2001. Synchro-nization and assembly formation in the visual cortex. Prog.Brain Res. 130: 111140

    Frien, A., R. Eckhorn, R. Bauer, T. Woelbern and H. Kehr. 1994.Stimulus specific oscillations at zero phase between visualareas V1 and V2 of awake monkey. Neuro Report 5:22732277.

    Gallant, J. L., J. Braun and D. C. Van Essen. 1993. Selectivity forpolar, hyperbolic, and cartesian gratings in macaque visualcortex. Science 259: 100103.

    Georgopoulos, A. P. 1990. Neural coding of the direction ofreaching and a comparison with saccadic eye movements.Cold Spring Harbor Laboratory Press 55: 849859.

    Gerstein, G. L. and D. H. Perkel. 1972. Mutual temporal relation-ships among neuronal spike trains. Bioph. J. 12: 453473.

    Gerstein, G. L., P. Bedenbaugh and A. M. H. J. Aertsen. 1989.Neuronal assemblies. IEEE Trans. Bio-Med. Eng. 36: 414.

    Gochin, P. M., E. K. Miller, C. G. Gross and G. L. Gerstein. 1991.

    Functonal interactions among neurons in inferior temporal cor-tex of the awake macaque. Exp. Brain Res. 84: 505516.

    Grammont, F. and A. Riehle. 1999. Precise spike synchronizationin monkey motor cortex involved in preparation for move-ment. Exp. Brain Res. 128: 118122.

    Gross, C. G., C. E. Rocha-Miranda and D. B. Bender. 1972. Vi-sual properties of neurons in inferotemporal cortex of themacaque. J. Neurophysiol. 35: 96111.

    Grossberg, S. 1980. How does a brain build a cognitive code?Psychol. Rev. 87: 151.

    Grossberg, S. and D. Somers. 1991. Synchronized oscillationsduring cooperative feature linking in a cortical model of visualperception. Neural Networks 4: 453466.

    Gruber, T., M. M. Mller, A. Keil and T. Elbert. 1999. Selectivevisual-spatial attention alters induced gamma band responsesin the human EEG. Clin. Neurophysiol. 110: 20742085.

    Hatsopoulos, N. G., C. L. Ojakangas, L. Paninski and J. P.Donoghue. 1998. Information about movement direction ob-tained from synchronous activity of motor cortical neurons.Proc. Nat. Acad. Sci. USA 95: 1570615711.

    Hebb, D. O. 1949. The Organization of Behavior. Wiley, NewYork.

    Henry, G. H. 1985. Physiology of cat striate cortex. In: CerebralCortex (A. Peters and E. G. Jones, eds.). Plenum Press, NewYork, pp. 119155.

    Hopfield, J. J. 1982. Neural networks and physical systems withemergent collective computational abilities. Proc. Natl. Acad.Sci. USA. 79: 25542558.

    Horn, D. and M. Usher. 1991. Parallel activation of memories inan oscillatory neural network. Neural Comp. 3: 3143.

    Hubel, D. H. and T. N. Wiesel. 1959. Receptive fields of singleneurons in the cat striate cortex. J. Physiol. 148: 574591.

    Hubel, D. H. and T. N. Wiesel. 1962. Receptive fields, binocularinteraction and functional architecture in the cats visual cor-tex. J. Physiol. 160: 106154.

    Janssen, P., R. Vogels and G. A. Orban. 1999. Macaque inferiortemporal neurons are selective for disparity-defined three-di-mensional shapes. Proc. Nat. Acad. Sci. USA 96: 82178222.

    Janssen, P., R. Vogels and G. A. Orban. 2000. Selectivity for 3Dshape that reveals distinct areas within macaque inferior tem-poral cortex. Science 288: 20542056.

    Knig, P. and T. B. Schillen. 1990. Segregation of oscillatory re-sponses by conflicting stimuli- desynchronizing connections inneural oscillator layers. In: Parallel processing in neural sys-tems and computers (R. Eckmiller, G. Hartmann and G.Hauske, eds.). Elsevier Science Publishers B.V., North Hol-land, pp. 117120.

    Kreiter, A. K. 1992. Kodierung neuronaler Assemblies durch ko-hrente Aktivitt: Korrelationsanalysen im Sehsystem von

    Sugetieren. University of Tuebingen.Kreiter, A. K. and W. Singer. 1992. Oscillatory neuronal re-sponses in the visual cortex of the awake macaque monkey.Eur. J. Neurosci. 4: 369375.

    Kreiter, A. K. and W. Singer. 1996a. On the role of neural syn-chrony in the primate visual cortex. In: BrainTheory (A. Aert-sen and V. Braitenberg, eds.). Elsevier, Amsterdam, pp.201227.

    Kreiter, A. K. and W. Singer. 1996b. Stimulus dependent syn-chronization of neuronal responses in the visual cortex of theawake macaque monkey. J. Neurosci. 16: 23812396.

    Krger, J. 1990. Multi-microelectrode investigation of monkeystriate cortex: link between correlational and neuronal proper-ties in the infragranular layers. Vis. Neurosci. 5: 135142.

    Livingstone, M. S. 1996. Oscillatory firing and interneuronal cor-

    relations in squirrel monkey striate cortex. J. Neurophysiol. 75:24672485.

    253

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    14/15

    Livingstone, M. S. and D. H. Hubel. 1988. Segregation of form,color, movement, and depth: Anatomy, physiology, and per-

    ception. Science 240: 740749.

    Logothetis, N. K., J. Pauls and T. Poggio. 1995. Shape represen-tation in the inferior temporal cortex of monkeys. Current Biol.5: 552563.

    Lutzenberger, W., F. Pulvermller, T. Elbert and N. Bierbaumer.1995. Visual stimulation alters local 40-Hz responses in hu-

    mans: an EEGstudy. Neurosci. Lett. 183: 3942.Maunsell, J. H. R. and W. T. Newsome. 1987. Visual processing

    in monkey extrastriate cortex. Ann. Rev. Neurosci. 10:363401.

    McClelland, J. L. and D. E. Rumelhart. 1986. Parallel Distributed

    Processing: Explorations in the Microstructure of Cognition.Volume 2: Psychological and Biological Models. MIT Press,

    Cambridge.Milner, P. M. 1974. A model for visual shape recognition. Psy-

    chol. Rev. 81: 521535.Miltner, W. H. R., C. Braun, M. Arnold, H. Witte and E. Taub.

    1999. Coherence of gamma-band activity as a basis for asso-ciative learning. Nature 397: 434436.

    Missal, M., R. Vogels, C. Y. Li and G. A. Orban. 1999. Shape in-teractions in macaque inferior temporal neurons. J. Neurophy-siol. 82: 131142.

    Miyashita, Y. 1993. Inferior temporal cortex: where visual per-ception meets memory. Ann. Rev. Neurosci. 16: 245263.

    Moore, G. P., J. P. Segundo, D. H. Perkel and H. Levitan. 1970.Statistical signs of synaptic interaction in neurons. Bioph. J.

    10: 876900.Munk, M. H. J., P. R. Roelfsema, P. Fries, A. K. Kreiter and W.

    Singer. 2000. Rapidly changing synchronisation of gamma-frequency oscillations across visual, parietal and motor areas

    of macaque monkeys performing a visuo-motor task. Soc.Neurosc. Abstr. 26: 20852085.

    Murthy, V. N. and E. E. Fetz. 1992. Coherent 25- to 35-Hz oscil-

    lations in the sensorimotor cortex of awake behaving monkeys.Proc. Natl. Acad. Sci. USA. 89: 56705674.

    Mller, M. M., J. Bosch, T. Elbert, A. K. Kreiter, M. Valdes Sosa,P. Valdes Sosa and B. Rockstroh. 1996. Visually induced

    gamma-band responses in human electroencephalographic ac-tivity a link to animal studies. Exp. Brain Res. 112: 96102.

    Neven, H. and A. Aertsen. 1992. Rate coherence and event coher-ence in the visual cortex: a neuronal model of object recogni-tion. Biol. Cybern. 67: 309322.

    Niebur, E. and C. Koch. 1994. A model for the neuronal imple-mentation of selective visual attention based on temporal cor-

    relation among neurons. J. Comput. Neurosci. 12: 141158.

    Niebur, E., C. Koch and C. Rosin. 1993. An oscillation-basedmodel for the neuronal basis of attention. Vision Res. 18:27892802.

    Nowak, L. G., M. H. Munk, A. C. James, P. Girard and J. Bullier.1999. Cross-correlation study of the temporal interactions be-tween areas V1 and V2 of the macaque monkey. J. Neurophys-

    iol. 81: 10571074.Olshausen, B. A., C. H. Anderson and D. C. Van Essen. 1993. A

    neurobiological model of visual attention and invariant patternrecognition based on dynamic routing of information. J. Neu-

    rosci. 13: 47004719.Orban, G. A. 1984. Neuronal operations in the visual cortex.

    Springer-Verlag, Berlin.Palm, G. 1982. Neural Assemblies. Springer, Heidelberg.

    Palm, G. 1990. Cell assemblies as a guideline for brain research.Concepts in Neurosci. 1: 133147.

    Perkel, D. H., G. L. Gerstein and G. P. Moore. 1967. Neuronalspike trains and stochastic point processes II. simultaneousspike trains. Bioph. J. 7: 419440.

    Perrett, D. I., P. A. J. Smith, D. D. Potter, A. J. Mistlin, A. S.Head, A. D. Milner and M. A. Jeeves. 1985. Visual cells in thetemporal cortex sensitive to face view and gaze direction. Proc.R. Soc. Lond. B 223: 293317.

    Revonsuo, A., M. Wilenius-Emet, J. Kuusela and M. Lehto.1997. The neural generation of a unified illusion in human vi-sion. Neuro Report 8: 38673870.

    Riehle, A., S. Grn, M. Diesmann and A. Aertsen. 1997. Spikesynchronization and rate modulation differentially involved inmotor cortical function. Science 278: 19501953.

    Ritz, R., W. Gerstner, U. Fuentes and J. L. van Hemmen. 1994. Abiologically motivated and analytically soluble model of col-lective oscillations in the cortex: II. Application to binding andpattern segmentation. Biol. Cybern. 71: 349359.

    Rodriguez, E., N. George, J. P. Lachaux, J. Martinerie, B. Renault

    and F. J. Varela. 1999. Perceptions shadow: long distancesynchronization of human brain activity. Nature 397: 430433.

    Rolls, E. T. 1992. Neurophysiological mechanisms underlyingface processing within and beyond the temporal cortical visualareas. Phil. Trans. R. Soc. Lond. B 335: 1121.

    Rumelhart, D. E. and J. L. McClelland. 1986. Parallel DistributedProcessing: Explorations in the Microstructure of Cognition.Volume 1: Foundations. MIT Press, Cambridge.

    Saito, H.-A., M. Yukie, K. Tanaka, K. Hikosaka, Y. Fukada andE. Iwai. 1986. Integration of direction signals of image motionin the superior temporal sulcus of the macaque monkey. J.Neurosci. 6: 145157.

    Sakai, K. and Y. Miyashita. 1991. Neural organization for thelong-term memory of paired associates. Nature 354: 152155.

    Sanes, J. N. and J. P. Donoghue. 1993. Oscillations in local field

    potentials of the primate motor cortex during voluntary move-ment. Proc. Natl. Acad. Sci. USA. 90: 44704474.

    Schillen, T. B. and P. Knig. 1994. Binding by temporal structurein multiple feature domains of an oscillatory neuronal network.Biol. Cybern. 70: 397405.

    Singer, W. 1990a. Search for coherence: a basic principle of corti-cal self-organization. Concepts in Neurosci. 1: 126.

    Singer, W. 1990b. The formation of cooperative cell assembliesin the visual cortex. J. Exp. Biol. 153: 177197.

    Singer, W. 1999. Neuronal synchrony: A versatile code for thedefinition of relations? Neuron 24: 4965.

    Singer, W. and C. M. Gray. 1995. Visual feature integration andthe temporal correlation hypothesis. Ann. Rev. Neurosci. 18:555586.

    Sompolinsky, H. and M. Tsodyks. 1994. Segmentation by a net-

    work of oscillators with stored memories. Neural Comp. 6:642657.Sporns, O., G. Tononi and G. M. Edelman. 1991. Modeling per-

    ceptual grouping and figure-ground segregation by means ofactive reentrant connections. Proc. Natl. Acad. Sci. USA. 88:129133.

    Steinmetz, P. N., A. Roy, P. J. Fitzgerald, S. S. Hsiao, K. O. John-son and E. Niebur. 2000. Attention modulates synchronizedneuronal firing in primate somatosensory cortex. Nature 404:187190.

    Tallon, C., O. Bertrand, P. Bouchet and J. Pernier. 1995. Gamma-range activity evoked by coherent visual stimuli in humans.Eur. J. Neurosci. 7: 12851291.

    Tallon-Baudry, C., O. Bertrand, C. Delpuech and J. Pernier.1996. Stimulus specificity of phase-locked and non-phase-

    locked 40 Hz visual responses in human. J. Neurosci. 16:42404249.

    254

    A. K. Kreiter

    Zoology 104 (2001) 34

  • 8/8/2019 Functional Implications of Temporal Structure in Primate Cortical Information Processing

    15/15

    Tallon-Baudry, C., O. Bertrand, C. Delpuech and J. Pernier.1997. Oscillatory gamma-band (3070 Hz) activity induced bya visual search task in humans. J. Neurosci. 17: 722734.

    Tallon-Baudry, C., O. Bertrand, F. Peronnet and J. Pernier. 1998.Induced gamma-band activity during the delay of a visualshort-term memory task in humans. J. Neurosci. 18:42444254.

    Tallon-Baudry, C., A. Kreiter and O. Bertrand. 1999. Sustainedand transient oscillatory responses in the gamma and betabands in a visual short-term memory task in humans. VisualNeurosci. 16: 449459.

    Tanaka, K. 1993. Neuronal mechanisms of object recognition.Science 262: 685688.

    Tanaka, K. 1996. Inferotemporal cortex and object vision. Ann.Rev. Neurosci. 19: 109139.

    Vaadia, E., I. Haalman, M. Abeles, H. P. Y. Bergman, H. Slovin andA. Aertsen. 1995. Dynamics of neuronal interactions in monkeycortex in relation to behavioural events. Nature 373: 515518.

    von der Malsburg, C. 1981. The correlation theory of brain func-tion. In: Internal Report 81-2 Department of Neurobiology,Max-Planck-Institute for Biophysical Chemistry, Gttingen.

    von der Malsburg, C. 1985. Nervous structures with dynamicallinks. Ber. Bunsenges. Phys. Chem. 89: 703710.

    von der Malsburg, C. 1986. Am I thinking assemblies? In: BrainTheory (G. Palm and A. Aertsen, eds.). Springer-Verlag,Berlin, pp. 161176.

    von der Malsburg, C. and W. Schneider. 1986. A neural cocktail-party processor. Biol. Cybern. 54: 2940.

    von der Malsburg, C. and W. Singer. 1988. Principles of corticalnetwork organization. In: Neurobiology of Neocortex (P.Rakic and W. Singer, eds.). John Wiley & Sons Limited,Chichester, pp. 6999.

    Wang, D., J. Buhmann and C. von der Malsburg. 1990.Pattern segmentation in associative memory. Neural Comp. 2:94106.

    Yamane, S., S. Kaji and K. Kawano. 1988. What facial featuresactivate face neurons in the inferotemporal cortex of the mon-key? Exp. Brain Res. 73: 209214.

    Young, M. P. and S. Yamane. 1992. Sparse population coding offaces in the inferotemporal cortex. Science 256: 13271331.

    Zeki, S. M. 1975. The functional organization of projections from

    striate to prestriate visual cortex in the rhesus monkey. ColdSpring Harbor Symp. Quant. Biol. 40: 591600.

    Zipser, D. and R. A. Andersen. 1988. A back-propagation pro-grammed network that simulates response properties of a sub-set of posterior parietal neurons. Nature 331: 679684.

    255

    Temporal structure in cortical information processing

    Zoology 104 (2001) 34