Title: Functions of the Primate Temporal Lobe Cortical Visual Areas in Invariant Visual Object and Face Recognition
Abstract: There is now good evidence that neural systems in temporal cortical visual areas process information about faces. Because a large number of neurons are devoted to this class of stimuli, these systems have proved amenable to experimental analysis. Face recognition and the identification of face expression are important in primate social behavior, and analysis of the neural systems involved is important for understanding the effects of damage to these systems in humans. Damage to these or related systems can lead to prosopagnosia, an impairment in recognizing individuals from the sight of their faces, or to difficulty in identifying the expression on a face. It turns out that the temporal cortical visual areas also have similar neuronal populations that code for objects, and study of both sets of neurons is helping to unravel the enormous computational problem of invariant visual object recognition. The neurophysiological recordings are made mainly in the macaque, a species of nonhuman primate, first because the temporal lobe, in which this processing occurs, is much more developed than in nonprimates; and second because the findings are relevant to understanding the effects of brain damage in patients. While recording in the temporal lobe cortical visual areas of macaques, Charles Gross and colleagues found some neurons that appeared to respond best to complex visual stimuli such as faces (16Desimone R. Gross C.G. Visual areas in the temporal lobe of the macaque.Brain Res. 1979; 178: 363-380Crossref PubMed Scopus (386) Google Scholar, 14Bruce C. Desimone R. Gross C.G. Visual properties of neurons in a polysensory area in superior temporal sulcus of the macaque.J. Neurophysiol. 1981; 46: 369-384Crossref PubMed Scopus (947) Google Scholar; see also 15Desimone R. Face-selective cells in the temporal cortex of monkeys.J. Cogn. Neurosci. 1991; 3: 1-8Crossref PubMed Scopus (387) Google Scholar). It was soon found that while some of these neurons could respond to parts of faces, other neurons required several parts of the face to be present in the correct spatial arrangement, and that many of these neurons did not just respond to any face that was shown but responded differently to different faces (56Perrett D.I. Rolls E.T. Caan W. Visual neurons responsive to faces in the monkey temporal cortex.Exp. Brain Res. 1982; 47: 329-342Crossref PubMed Scopus (921) Google Scholar, 17Desimone R. Albright T.D. Gross C.G. Bruce C. Stimulus-selective properties of inferior temporal neurons in the macaque.J. Neurosci. 1984; 4: 2051-2062Crossref PubMed Google Scholar, 63Rolls E.T. Neurons in the cortex of the temporal lobe and in the amygdala of the monkey with responses selective for faces.Hum. Neurobiol. 1984; 3: 209-222PubMed Google Scholar, 29Gross C.G. Desimone R. Albright T.D. Schwartz E.L. Inferior temporal cortex and pattern recognition.Exp. Brain. Res. Suppl. 1985; 11: 179-201Crossref Google Scholar). By responding differently to different faces, these neurons potentially encode information useful for identifying individual faces. It also appears that there is some specialization of function of different temporal cortical visual areas, as described next. The visual pathways project from the primary visual cortex to the temporal lobe visual cortical areas by a number of intervening cortical stages (89Seltzer B. Pandya D.N. Afferent cortical connections and architectonics of the superior temporal sulcus and surrounding cortex in the rhesus monkey.Brain Res. 1978; 149: 1-24Crossref PubMed Scopus (580) Google Scholar, 44Maunsell J.H.R. Newsome W.T. Visual processing in monkey extrastriate cortex.Annu. Rev. Neurosci. 1987; 10: 363-401Crossref PubMed Scopus (763) Google Scholar, 3Baizer J.S. Ungerleider L.G. Desimone R. Organization of visual inputs to the inferior temporal and posterior parietal cortex in macaques.J. Neurosci. 1991; 11: 168-190Crossref PubMed Google Scholar). The inferior temporal visual cortex, area TE, is divided on the basis of cytoarchitecture, myeloarchitecture, and afferent input into areas TEa, TEm, TE3, TE2, and TE1. In addition there is a set of different areas in the cortex in the superior temporal sulcus (89Seltzer B. Pandya D.N. Afferent cortical connections and architectonics of the superior temporal sulcus and surrounding cortex in the rhesus monkey.Brain Res. 1978; 149: 1-24Crossref PubMed Scopus (580) Google Scholar, 8Baylis G.C. Rolls E.T. Leonard C.M. Functional subdivisions of temporal lobe neocortex.J. Neurosci. 1987; 7: 330-342PubMed Google Scholar) (see Figure 1). Of these latter areas, TPO receives inputs from temporal, parietal, and occipital cortex; PGa and IPa from parietal and temporal cortex; and TS and TAa primarily from auditory areas (89Seltzer B. Pandya D.N. Afferent cortical connections and architectonics of the superior temporal sulcus and surrounding cortex in the rhesus monkey.Brain Res. 1978; 149: 1-24Crossref PubMed Scopus (580) Google Scholar). There is considerable specialization of function in these architectonically defined areas (8Baylis G.C. Rolls E.T. Leonard C.M. Functional subdivisions of temporal lobe neocortex.J. Neurosci. 1987; 7: 330-342PubMed Google Scholar). Areas TPO, PGa, and IPa are multimodal, with neurons that respond to visual, auditory, and/or somatosensory inputs. The more ventral areas in the inferior temporal gyrus (areas TE3, TE2, TE1, TEa, and TEm) are primarily unimodal visual areas. Areas in the cortex in the anterior and dorsal part of the superior temporal sulcus (e.g., TPO, IPa, and IPg) have neurons specialized for the analysis of moving visual stimuli. Neurons responsive primarily to faces are found more frequently in areas TPO, TEa, and TEm, where they comprise ∼20% of the visual neurons responsive to stationary stimuli, in contrast to the other temporal cortical areas, where they comprise 4%–10%. The stimuli that activate other cells in these TE regions include simple visual patterns such as gratings and combinations of simple stimulus features (29Gross C.G. Desimone R. Albright T.D. Schwartz E.L. Inferior temporal cortex and pattern recognition.Exp. Brain. Res. Suppl. 1985; 11: 179-201Crossref Google Scholar, 94Tanaka K. Saito C. Fukada Y. Moriya M. Integration of form, texture, and color information in the inferotemporal cortex of the macaque.in: Iwai E. Mishkin M. Vision, Memory and the Temporal Lobe. Elsevier, New York1990Google Scholar). Due to the fact that face-selective neurons have a wide distribution, it might be expected that only large lesions, or lesions that interrupt outputs of these visual areas, would produce readily apparent face-processing deficits. Moreover, neurons with responses related to facial expression, movement, and gesture are more likely to be found in the cortex in the superior temporal sulcus, whereas neurons with activity related to facial identity are more likely to be found in the TE areas (see below and 30Hasselmo M.E. Rolls E.T. Baylis G.C. The role of expression and identity in the face-selective responses of neurons in the temporal visual cortex of the monkey.Behav. Brain Res. 1989; 32 (a): 203-218Crossref PubMed Scopus (505) Google Scholar). Neurons with responses selective for faces respond 2–20 times more to faces than to a wide range of gratings, simple geometrical stimuli, or complex 3D objects (see 63Rolls E.T. Neurons in the cortex of the temporal lobe and in the amygdala of the monkey with responses selective for faces.Hum. Neurobiol. 1984; 3: 209-222PubMed Google Scholar, 68Rolls E.T. Neurophysiological mechanisms underlying face processing within and beyond the temporal cortical visual areas.Philos. Trans. R. Soc. Lond. B Biol. Sci. 1992; 335 (b): 11-21Crossref PubMed Scopus (287) Google Scholar, 7Baylis G.C. Rolls E.T. Leonard C.M. Selectivity between faces in the responses of a population of neurons in the cortex in the superior temporal sulcus of the monkey.Brain Res. 1985; 342: 91-102Crossref PubMed Scopus (225) Google Scholar, 8Baylis G.C. Rolls E.T. Leonard C.M. Functional subdivisions of temporal lobe neocortex.J. Neurosci. 1987; 7: 330-342PubMed Google Scholar). The responses to faces are excitatory, with firing rates often reaching 100 spikes/s, and sustained, and they have typical latencies of 80–100 ms. The neurons are typically unresponsive to auditory or tactile stimuli and to the sight of arousing or aversive stimuli. These findings indicate that explanations in terms of arousal, emotional or motor reactions, and simple visual feature sensitivity are insufficient to account for the selective responses to faces and face features observed in this population of neurons (56Perrett D.I. Rolls E.T. Caan W. Visual neurons responsive to faces in the monkey temporal cortex.Exp. Brain Res. 1982; 47: 329-342Crossref PubMed Scopus (921) Google Scholar, 7Baylis G.C. Rolls E.T. Leonard C.M. Selectivity between faces in the responses of a population of neurons in the cortex in the superior temporal sulcus of the monkey.Brain Res. 1985; 342: 91-102Crossref PubMed Scopus (225) Google Scholar, 72Rolls E.T. Baylis G.C. Size and contrast have only small effects on the responses to faces of neurons in the cortex of the superior temporal sulcus of the monkey.Exp. Brain Res. 1986; 65: 38-48Crossref PubMed Scopus (202) Google Scholar). Observations consistent with these findings have been published by 17Desimone R. Albright T.D. Gross C.G. Bruce C. Stimulus-selective properties of inferior temporal neurons in the macaque.J. Neurosci. 1984; 4: 2051-2062Crossref PubMed Google Scholar, who described a similar population of neurons located primarily in the cortex in the superior temporal sulcus, which responded to faces but not to simpler stimuli such as edges and bars or to complex nonface stimuli (see also 29Gross C.G. Desimone R. Albright T.D. Schwartz E.L. Inferior temporal cortex and pattern recognition.Exp. Brain. Res. Suppl. 1985; 11: 179-201Crossref Google Scholar). These neurons are specialized to provide information about faces in that they provide much more information (on average 0.4 bits) about which (of 20) face stimuli are being seen than about which (of 20) nonface stimuli are being seen (on average 0.07 bits) (75Rolls E.T. Tovee M.J. Sparseness of the neuronal representation of stimuli in the primate temporal visual cortex.J. Neurophysiol. 1995; 73 (a): 713-726Crossref PubMed Scopus (365) Google Scholar, 85Rolls E.T. Treves A. Tovee M.J. The representational capacity of the distributed encoding of information provided by populations of neurons in the primate temporal visual cortex.Exp. Brain Res. 1997; 114: 149-162Crossref PubMed Scopus (167) Google Scholar). These information theoretic procedures provide an objective and quantitative way to show what is "represented" by a particular population of neurons. Masking out or presenting parts of the face (e.g., eyes, mouth, or hair) in isolation reveals that different cells respond to different features or subsets of features. For some cells, responses to the normal organization of cut-out or line-drawn facial features are significantly larger than to images in which the same facial features are jumbled (56Perrett D.I. Rolls E.T. Caan W. Visual neurons responsive to faces in the monkey temporal cortex.Exp. Brain Res. 1982; 47: 329-342Crossref PubMed Scopus (921) Google Scholar, 82Rolls E.T. Tovee M.J. Purcell D.G. Stewart A.L. Azzopardi P. The responses of neurons in the temporal cortex of primates, and face identification and detection.Exp. Brain Res. 1994; 101 (a): 474-484Crossref Scopus (84) Google Scholar). These findings are consistent with the hypotheses developed below that by competitive self-organization some neurons in these regions respond to parts of faces by responding to combinations of simpler visual properties received from earlier stages of visual processing, and that other neurons respond to combinations of parts of faces and thus respond only to whole faces. Moreover, the finding that for some of these latter neurons the parts must be in the correct spatial configuration shows that the combinations formed can reflect not just the features present but also their spatial arrangement. This provides a way in which binding can be implemented in neural networks (see 21Elliffe M.C.M. Rolls E.T. Stringer S.M. Invariant recognition of feature combinations in the visual system.Biol. Cybern. 2000; in press (b)Google Scholar). Further evidence that neurons in these regions respond to combinations of features in the correct spatial configuration was found by Tanaka et al. (e.g., 1990), using combinations of features that are used by comparable neurons to define objects. An important question for understanding brain function is whether a particular object (or face) is represented in the brain by the firing of one or a few (gnostic or "grandmother") cells (5Barlow H.B. Single units and sensation a neuron doctrine for perceptual psychology?.Perception. 1972; 1: 371-394Crossref PubMed Scopus (1047) Google Scholar), or whether instead the firing of a group or ensemble of cells each with different profiles of responsiveness to the stimuli provides the representation. A grandmother cell representation is a code which is very sparse, in that each neuron responds to only one object or stimulus. A very large number of neurons would be required, since each neuron responds to only one stimulus. This encoding is described as local, in that all the information that a particular object is present is carried by one neuron. In contrast, ensemble encoding is described as distributed, in that the information that a particular stimulus was shown is distributed across a population of neurons. Many more stimuli can potentially be represented by a distributed code, as each object is represented by a combination of different neurons firing, and this type of code can have many other advantages, as described below. The actual representation found is distributed. 7Baylis G.C. Rolls E.T. Leonard C.M. Selectivity between faces in the responses of a population of neurons in the cortex in the superior temporal sulcus of the monkey.Brain Res. 1985; 342: 91-102Crossref PubMed Scopus (225) Google Scholar showed this with the responses of temporal cortical neurons that typically responded to several members of a set of 5 faces, with each neuron having a different profile of responses to each face. In a more recent study using 23 faces and 45 nonface natural images, a distributed representation was found again (75Rolls E.T. Tovee M.J. Sparseness of the neuronal representation of stimuli in the primate temporal visual cortex.J. Neurophysiol. 1995; 73 (a): 713-726Crossref PubMed Scopus (365) Google Scholar), with the average sparseness being 0.65. (The sparseness of the representation provided by a neuron can be defined as where rs is the mean firing rate of the neuron to stimulus s in the set of S stimuli [see Rolls and Treves, 1998]. If the neurons are binary [either firing or not to a given stimulus], then a would be 0.5 if the neuron responded to 50% of the stimuli, and 0.1 if a neuron responded to 10% of the stimuli.) If the spontaneous firing rate was subtracted from the firing rate of the neuron to each stimulus, so that the changes of firing rate, i.e., the active responses of the neurons, were used in the sparseness calculation, then the "response sparseness" had a lower value, with a mean of 0.33 for the population of neurons. The distributed nature of the representation can be further understood by the finding that the firing rate distribution of single neurons when a wide range of natural visual stimuli are being viewed is approximately exponentially distributed, with rather few stimuli producing high firing rates, and increasingly large numbers of stimuli producing lower and lower firing rates (75Rolls E.T. Tovee M.J. Sparseness of the neuronal representation of stimuli in the primate temporal visual cortex.J. Neurophysiol. 1995; 73 (a): 713-726Crossref PubMed Scopus (365) Google Scholar, 2Baddeley R.J. Abbott L.F. Booth M.J.A. Sengpiel F. Freeman T. Wakeman E.A. Rolls E.T. Responses of neurons in primary and inferior temporal visual cortices to natural scenes.Proc. R. Soc. Lond. B Biol. Sci. 1997; 264: 1775-1783Crossref PubMed Scopus (312) Google Scholar, 104Treves A. Panzeri S. Rolls E.T. Booth M. Wakeman E.A. Firing rate distributions and efficiency of information transmission of inferior temporal cortex neurons to natural visual stimuli.Neural Comput. 1999; 11: 611-641Crossref Scopus (65) Google Scholar) (see Figure 2). This is a clear answer to whether these neurons are grandmother cells: they are not, in the sense that each neuron has a graded set of responses to the different members of a set of stimuli, with the prototypical distribution similar to that of the neuron illustrated in Figure 2. On the other hand, each neuron does respond very much more to some stimuli than to many others and in this sense is tuned to some stimuli. The sparseness of such an exponential distribution of firing rates is 0.5. It has been shown that the distribution may arise from the threshold nonlinearity of neurons combined with short-term variability in the responses of neurons (104Treves A. Panzeri S. Rolls E.T. Booth M. Wakeman E.A. Firing rate distributions and efficiency of information transmission of inferior temporal cortex neurons to natural visual stimuli.Neural Comput. 1999; 11: 611-641Crossref Scopus (65) Google Scholar). The distributed properties of the code used are further revealed by applying information theory (see 90Shannon C.E. A mathematical theory of communication.ATT Bell Labs. Tech. J. 1948; 27: 379-428Crossref Scopus (20769) Google Scholar, 43MacKay D.M. McCullough W.S. The limiting information capacity of a neuronal link.Bull. Math. Biophys. 1952; 14: 127-135Crossref Scopus (202) Google Scholar, 19Eckhorn R. Popel B. Rigorous and extended application of information theory to the afferent visual system of the cat.Biol. Cybern. 1974; 16: 191-200Google Scholar, 78Rolls E.T. Treves A Neural Networks and Brain Function. Oxford University Press, Oxford1998Google Scholar [Appendix 2]) to analyze how information is represented by a population of these neurons. The information required to identify which of S equiprobable stimuli were shown is log2S bits. (Thus, one bit is required to specify which of two stimuli was shown, two bits to specify which of four stimuli was shown, three bits to specify which of eight stimuli was shown, etc.) If the encoding was local (or grandmother cell–like), the number of stimuli encoded by a population of neurons would be expected to rise approximately linearly with the number of neurons in the population. In contrast, with distributed encoding, provided that the neuronal responses are sufficiently independent and reliable (not too noisy), the number of stimuli encodable by the population of neurons might be expected to rise exponentially as the number of neurons in the sample of the population was increased. The information about which of 20 equiprobable faces had been shown that was available from the responses of different numbers of these neurons is shown in Figure 3. First, it is clear (Figure 3) that the information rises approximately linearly, and the number of stimuli encoded thus rises approximately exponentially, as the number of cells in the sample increases (1Abbott L.F. Rolls E.T. Tovee M.J. Representational capacity of face coding in monkeys.Cereb. Cortex. 1996; 6: 498-505Crossref PubMed Scopus (82) Google Scholar, 85Rolls E.T. Treves A. Tovee M.J. The representational capacity of the distributed encoding of information provided by populations of neurons in the primate temporal visual cortex.Exp. Brain Res. 1997; 114: 149-162Crossref PubMed Scopus (167) Google Scholar; see also 78Rolls E.T. Treves A Neural Networks and Brain Function. Oxford University Press, Oxford1998Google Scholar). This direct neurophysiological evidence thus demonstrates that the encoding is distributed, and the responses are sufficiently independent and reliable, such that the representational capacity increases exponentially with the number of neurons in the ensemble. The consequence of this is that large numbers of stimuli, and fine discriminations between them, can be represented without having to measure the activity of an enormous number of neurons. (It has been shown that the main reason why the information tends to asymptote, as shown in Figure 3, as the number of neurons in the sample increases is just that the ceiling is being approached of how much information is required to discriminate between the sets of stimuli, which with 20 stimuli is log220 = 4.32 bits [1Abbott L.F. Rolls E.T. Tovee M.J. Representational capacity of face coding in monkeys.Cereb. Cortex. 1996; 6: 498-505Crossref PubMed Scopus (82) Google Scholar].) Second, it is clear that some information is available from the responses of just one neuron—on average, ∼0.34 bits. Thus, knowing the activity of just one neuron in the population does provide some evidence about which stimulus was present, even when the activity of all the other neurons is not known. This indicates that information is made explicit in the firing of individual neurons in a way that will allow neuronally plausible decoding, in which a receiving neuron simply uses each of its synaptic strengths to weight the input activity being received from each afferent axon and sums the result over all inputs (see below).Figure 3The Average Information from Different Numbers of Inferior Temporal Cortex Neurons about which of 20 Faces Had Been ShownShow full caption(a) The values for the average information available in the responses of different numbers of these neurons on each trial, about which of a set of 20 face stimuli has been shown. The decoding method was DP (diamonds) or PE (crosses), and the effects obtained with cross-validation procedures utilizing 50% of the trials as test trials are shown. The remainder of the trials in the cross-validation procedure were used as training trials. The full line indicates the amount of information expected from populations of increasing size, when assuming random correlations within the constraint given by the ceiling (the information in the stimulus set; I = 4.32 bits).(b) The percent correct for the corresponding data to those shown in Figure 3A. After Rolls et al. 1997.View Large Image Figure ViewerDownload Hi-res image Download (PPT) (a) The values for the average information available in the responses of different numbers of these neurons on each trial, about which of a set of 20 face stimuli has been shown. The decoding method was DP (diamonds) or PE (crosses), and the effects obtained with cross-validation procedures utilizing 50% of the trials as test trials are shown. The remainder of the trials in the cross-validation procedure were used as training trials. The full line indicates the amount of information expected from populations of increasing size, when assuming random correlations within the constraint given by the ceiling (the information in the stimulus set; I = 4.32 bits). (b) The percent correct for the corresponding data to those shown in Figure 3A. After Rolls et al. 1997. It has recently been shown that there are neurons in the inferior temporal visual cortex that encode view-invariant representations of objects, and for these neurons the same type of representation is found, namely distributed encoding with independent information conveyed by different neurons (11Booth M.C.A. Rolls E.T. View-invariant representations of familiar objects by neurons in the inferior temporal visual cortex.Cereb. Cortex. 1998; 8: 510-523Crossref PubMed Scopus (278) Google Scholar). The analyses just described were obtained with neurons that were not simultaneously recorded, but similar results have now been obtained with simultaneously recorded neurons—that is, the information about which stimulus was shown increases approximately linearly with the number of neurons, showing that the neurons convey information that is nearly independent (53Panzeri S. Schultz S.R. Treves A. Rolls E.T. Correlations and the encoding of information in the nervous system.Proc. R. Soc. B Biol. Sci. 1999; 266 (b): 1001-1012Crossref PubMed Scopus (229) Google Scholar, 87Rolls E.T. Tovee M.J. Panzeri S. The neurophysiology of backward visual masking information analysis.J. Cogn. Neurosci. 1999; 11: 335-346Crossref Scopus (163) Google Scholar, Soc. Neurosci., abstract). (Consistently28Gawne T.J. Richmond B.J. How independent are the messages carried by adjacent inferior temporal cortical neurons?.J. Neurosci. 1993; 13: 2758-2771Crossref PubMed Google Scholar showed that even adjacent pairs of neurons recorded simultaneously from the same electrode carried information that was ∼80% independent.) 53Panzeri S. Schultz S.R. Treves A. Rolls E.T. Correlations and the encoding of information in the nervous system.Proc. R. Soc. B Biol. Sci. 1999; 266 (b): 1001-1012Crossref PubMed Scopus (229) Google Scholar developed a method for measuring the information in the relative time of firing of simultaneously recorded neurons, which might be significant if the neurons became synchronized to some but not other stimuli in a set, as postulated by Singer and colleagues (e.g.22Engel A.K. Konig P. Kreiter A.K. Schillen T.B. Singer W. Temporal coding in the visual system new vistas on integration in the nervous system.Trends Neurosci. 1992; 15: 218-226Abstract Full Text PDF PubMed Scopus (499) Google Scholar). We found that for the set of inferior temporal cortex neurons currently available, almost all the information was available in the firing rates of the cells, and almost no information was available about which static image was shown in the relative time of firing of different simultaneously recorded neurons (53Panzeri S. Schultz S.R. Treves A. Rolls E.T. Correlations and the encoding of information in the nervous system.Proc. R. Soc. B Biol. Sci. 1999; 266 (b): 1001-1012Crossref PubMed Scopus (229) Google Scholar, 87Rolls E.T. Tovee M.J. Panzeri S. The neurophysiology of backward visual masking information analysis.J. Cogn. Neurosci. 1999; 11: 335-346Crossref Scopus (163) Google Scholar, Soc. Neurosci., abstract). Consistently, there were no significant cross-correlations between the spikes of these simultaneously recorded inferior temporal cortex neurons. Thus, the evidence is that most of the information is available in the firing rates of the neurons and not in synchronization for representations of faces and objects in the inferior temporal visual cortex (and this is also the case for space in the hippocampus and for odors in the orbitofrontal cortex; see 84Rolls E.T. Critchley H.D. Treves A. The representation of olfactory information in the primate orbitofrontal cortex.J. Neurophysiol. 1996; 75: 1982-1996PubMed Google Scholar). It is unlikely that there are further processing areas beyond those described where ensemble coding changes into grandmother cell (local) encoding. Anatomically, there does not appear to be a whole further set of visual processing areas present in the brain, and outputs from the temporal lobe visual areas such as those described are taken to limbic and related regions such as the amygdala, the orbitofrontal cortex, and—via the entorhinal cortex—the hippocampus, where associations between the visual stimuli and other sensory representations are formed (see 78Rolls E.T. Treves A Neural Networks and Brain Function. Oxford University Press, Oxford1998Google Scholar, 70Rolls E.T The Brain and Emotion. Oxford University Press, Oxford1999Google Scholar). Indeed, tracing this pathway onward, we have found a population of neurons with face-selective responses in the amygdala (40Leonard C.M. Rolls E.T. Wilson F.A.W. Baylis G.C. Neurons in the amygdala of the monkey with responses selective for faces.Behav. Brain Res. 1985; 15: 159-176Crossref PubMed Scopus (297) Google Scholar, 67Rolls E.T. Neurophysiology and functions of the primate amygdala.in: Aggleton J.P. The Amygdala. Wiley-Liss, New York1992Google Scholar, 71Rolls E.T. Neurophysiology and functions of the primate amygdala, and the neural basis of emotion.in: Aggleton J.P. The Amygdala A Functional Analysis. Oxford University Press, Oxford2000Google Scholar) and orbitofrontal cortex (Booth et al., 1998, Soc. Neurosci., abstract), and in the majority of these neurons different responses occur to different faces, with ensemble (not local) coding still being present. The amygdala in turn projects to another structure that may be important in other behavioral responses to faces, the ventral striatum, and comparable neurons have also been found in the ventral striatum (108Williams G.V. Rolls E.T. Leonard C.M. Stern C. Neuronal responses in the ventral striatum of the behaving macaque.Behav. Brain Res. 1993; 55: 243-252Crossref PubMed Scopus (114) Google Scholar). The advantages of the distributed encoding actually found are as follows. (These advantages do not apply to local, that is grandmother cell, encoding schemes, nor to all distributed encoding schemes [see 78Rolls E.T. Treves A Neural Networks and Brain Function. Oxford University Press, Oxford1998Google Scholar].) This property arises from two factors: (1) the encoding is sufficiently close to independent by the different neurons (i.e., factorial), and (2) the encoding is sufficiently distributed. Part of the biological significance of the exponential encoding capacity is that a receiving neuron or neurons can obtain information about which one of a very large number of stimuli is present by receiving the activity of relatively small numbers of in