Deep recurrent neural network reveals a hierarchy of process memory during dynamic natural vision

The human visual cortex extracts both spatial and temporal visual features to support perception and guide behavior. Deep convolutional neural networks (CNNs) provide a computational framework to model cortical representation and organization for spatial visual processing, but unable to explain how the brain processes temporal information. To overcome this limitation, we extended a CNN by adding recurrent connections to different layers of the CNN to allow spatial representations to be remembered and accumulated over time. The extended model, or the recurrent neural network (RNN), embodied a hierarchical and distributed model of process memory as an integral part of visual processing. Unlike the CNN, the RNN learned spatiotemporal features from videos to enable action recognition. The RNN better predicted cortical responses to natural movie stimuli than the CNN, at all visual areas, especially those along the dorsal stream. As a fully observable model of visual processing, the RNN also revealed a cortical hierarchy of temporal receptive window, dynamics of process memory, and spatiotemporal representations. These results support the hypothesis of process memory, and demonstrate the potential of using the RNN for in‐depth computational understanding of dynamic natural vision.     

Massive somatic deafferentation and motor deefferentation of the lower part of the body impair its visual recognition: a psychophysical study of patients with spinal cord injury

Embodied cognition theories postulate that perceiving and understanding the body states of other individuals are underpinned by the neural structures activated during first‐hand experience of the same states. This suggests that one’s own sensorimotor system may be used to identify the actions and sensations of others. Virtual and real brain lesion studies show that visual processing of body action and body form relies upon neural activity in the ventral premotor and the extrastriate body areas, respectively. We explored whether visual body perception may also be altered in the absence of damage to the above cortical regions by testing healthy controls and spinal cord injury (SCI) patients whose brain was unable to receive somatic information from and send motor commands to the lower limbs. Participants performed tasks investigating the ability to visually discriminate changes in the form or action of body parts affected by somatosensory and motor disconnection. SCI patients showed a specific, cross‐modal deficit in the visual recognition of the disconnected lower body parts. This deficit affected both body action and body form perception, hinting at a pervasive influence of ongoing body signals on the brain network dedicated to visual body processing. Testing SCI patients who did or did not practise sports allowed us to test the influence of motor practice on visual body recognition. We found better upper body action recognition in sport‐practising SCI patients, indicating that motor practice is useful for maintaining visual representation of actions after deafferentation and deefferentation. This may be a potential resource to be exploited for rehabilitation.     

Correlation between facial pattern recognition and brain composition in paper wasps

Unique among insects, some paper wasp species recognize conspecific facial patterns during social communication. To evaluate whether specialized brain structures are involved in this task, we measured brain and brain component size in four different paper wasp species, two of which show facial pattern recognition. These behavioral abilities were not reflected by an increase in brain size or an increase in the size of the primary visual centers (medulla, lobula). Instead, wasps showing face recognition abilities had smaller olfactory centers (antennal lobes). Although no single brain compartment explains the wasps' specialized visual abilities, multi-factorial analysis of the different brain components, particularly the antennal lobe and the mushroom body sub-compartments, clearly separates those species that show facial pattern recognition from those that do not. Thus, there appears to be some neural specialization for visual communication in Polistes. However, the apparent lack of optic lobe specialization suggests that the visual processing capabilities of paper wasps might be preadapted for pattern discrimination and the ability to discriminate facial markings could require relatively small changes in their neuronal substrate.     

Dynamical consequences of lesions in cortical networks

To understand the effects of a cortical lesion it is necessary to consider not only the loss of local neural function, but also the lesion-induced changes in the larger network of endogenous oscillatory interactions in the brain. To investigate how network embedding influences a region's functional role, and the consequences of its being damaged, we implement two models of oscillatory cortical interactions, both of which inherit their coupling architecture from the available anatomical connection data for macaque cerebral cortex. In the first model, node dynamics are governed by Kuramoto phase oscillator equations, and we investigate the sequence in which areas entrain one another in the transition to global synchrony. In the second model, node dynamics are governed by a more realistic neural mass model, and we assess long-run inter-regional interactions using a measure of directed information flow. Highly connected parietal and frontal areas are found to synchronize most rapidly, more so than equally highly connected visual and somatosensory areas, and this difference can be explained in terms of the network's clustered architecture. For both models, lesion effects extend beyond the immediate neighbors of the lesioned site, and the amplitude and dispersal of nonlocal effects are again influenced by cluster patterns in the network. Although the consequences of in vivo lesions will always depend on circuitry local to the damaged site, we conclude that lesions of parietal regions (especially areas 5 and 7a) and frontal regions (especially areas 46 and FEF) have the greatest potential to disrupt the integrative aspects of neocortical function.     

Hyperexcitatory activity in visual cortex in homonymous hemianopia after stroke.

OBJECTIVES: Damage to and destruction of neural afferents result in a disruption of sensory input, which causes reduced activity in the corresponding cortical areas. Conversely, there is also evidence that lesions in the sensory pathway induce changes in the intracortical connectivity resulting in augmented cortical activity due to disinhibition. As disinhibition is assumed to be involved in the reconfiguration of neural networks, its appearance after brain lesions might be relevant for the restitution of impaired brain functions. METHODS: The effects of lesions in the visual pathway on the activity in visual cortex were studied using magnetoencephalography. In order to compare the neural activity affected by the lesion with the activity associated with intact visual processing, only patients with unilateral, post-chiasmatic lesions resulting in homonymous hemianopia were examined. RESULTS: Stimulation within the scotoma resulted in reduced magnetic activity compared to the stimulation of the intact hemifield. Increased activity was observed when the border region of the scotoma was stimulated. CONCLUSIONS: It is concluded that the magnetic hyperactivity reflects cortical disinhibition induced by lesions in the visual system. Furthermore, the possible role of cortical disinhibition as a basis for cortical reorganization and as a precondition for the recovery of impaired visual functions is discussed.     

Effective connectivity of brain regions related to visual word recognition: An fMRI study of Chinese reading

Past neuroimaging studies have focused on identifying specialized functional brain systems for processing different components of reading, such as orthography, phonology, and semantics. More recently, a few experiments have been performed to look into the integration and interaction of distributed neural systems for visual word recognition, suggesting that lexical processing in alphabetic languages involves both ventral and dorsal neural pathways originating from the visual cortex. In the present functional magnetic resonance imaging study, we tested the multiple pathways model with Chinese character stimuli and examined how the neural systems interacted in reading Chinese. Using dynamic causal modeling, we demonstrated that visual word recognition in Chinese engages the ventral pathway from the visual cortex to the left ventral occipitotemporal cortex, but not the dorsal pathway from the visual cortex to the left parietal region. The ventral pathway, however, is linked to the superior parietal lobule and the left middle frontal gyrus (MFG) so that a dynamic neural network is formed, with information flowing from the visual cortex to the left ventral occipitotemporal cortex to the parietal lobule and then to the left MFG. The findings suggest that cortical dynamics is constrained by the differences in how visual orthographic symbols in writing systems are linked to spoken language. Hum Brain Mapp 36:2580–2591, 2015. © 2015 Wiley Periodicals, Inc .     

Recognition memory and the medial temporal lobe: From monkey research to human pathology

This review provides a historical overview of decades of research on recognition memory, the process that allows both humans and animals to tell familiar from novel items. The emphasis is put on how monkey research improved our understanding of the medial temporal lobe (MTL) role and how tasks designed for monkeys influenced research in humans. The story starts in the early 1950s. Back then, memory was not a fashionable scientific topic. It was viewed as a function of the whole brain and not of specialized brain areas. All that changed in 1957–1958 when Brenda Milner, a neuropsychologist from Montreal, described patient H.M. He forgot all events as he lived them despite a fully preserved intelligence. He had received a MTL resection to relieve epilepsy. H.M. (1926–2008) would become the most influential patient in brain science. Which structures among those included in H.M.’s large lesion were important for recognition memory could not be evaluated in humans. It was gradually understood only after the successful development of a monkey model of human amnesia by Mishkin in 1978. Selective lesions and two behavioral tasks, delayed nonmatching-to-sample and visual paired comparison, were used to distinguish the contribution of the hippocampus from that of adjacent cortical areas. Driven by findings in non-human primates, human research on recognition memory is now trying to solve the question of whether the different structures composing MTL contributes to familiarity and recollection, the two possible forms taken by recognition. We described in particular two French patients, FRG and JMG, whose deficits support the currently dominant model attributing to the perirhinal cortex a critical role in recognition memory. Research on recognition memory has implications for the clinician as it may help understanding the cognitive deficits observed in different diseases. An illustration of such approach, linking basic and applied research, is provided for Alzheimer's disease.     

How do children learn to follow gaze,share joint attention,imitate their teachers,and use tools during social interactions?

How does an infant learn through visual experience to imitate actions of adult teachers, despite the fact that the infant and adult view one another and the world from different perspectives? To accomplish this, an infant needs to learn how to share joint attention with adult teachers and to follow their gaze towards valued goal objects. The infant also needs to be capable of view-invariant object learning and recognition whereby it can carry out goal-directed behaviors, such as the use of tools, using different object views than the ones that its teachers use. Such capabilities are often attributed to “mirror neurons”. This attribution does not, however, explain the brain processes whereby these competences arise. This article describes the CRIB (Circular Reactions for Imitative Behavior) neural model of how the brain achieves these goals through inter-personal circular reactions. Inter-personal circular reactions generalize the intra-personal circular reactions of Piaget, which clarify how infants learn from their own babbled arm movements and reactive eye movements how to carry out volitional reaches, with or without tools, towards valued goal objects. The article proposes how intra-personal circular reactions create a foundation for inter-personal circular reactions when infants and other learners interact with external teachers in space. Both types of circular reactions involve learned coordinate transformations between body-centered arm movement commands and retinotopic visual feedback, and coordination of processes within and between the What and Where cortical processing streams. Specific breakdowns of model processes generate formal symptoms similar to clinical symptoms of autism.     

Sensory Acquisition in the Cerebellum: An fMRI Study of Cerebrocerebellar Interaction During Visual Duration Discrimination

It has been suggested that the cerebellum participates in diverse neuropsychological functions by adjusting the sensory information acquired for the connected brain regions to support its processing capabilities. Nevertheless, the knowledge of how the cerebellum is modulated by the sensory information is far from clear. Function magnetic resonance imaging was exploited to investigate how the cerebellum activity and cerebrocerebellum interaction can be affected by the interaction between visual size and duration information during visual duration discrimination. The present findings support the sensory acquisition hypothesis that the cerebellum, together with extensive cortical networks, yields higher activation with incongruent sensory information to cope with increasing cortical computational demand. Furthermore, comprehensive intracerebellum connections are engaged in tasks with congruent sensory information for saving cortical computation with integrated sensory information.     

How does the brain rapidly learn and reorganize view-invariant and position-invariant object representations in the inferotemporal cortex?

All primates depend for their survival on being able to rapidly learn about and recognize objects. Objects may be visually detected at multiple positions, sizes, and viewpoints. How does the brain rapidly learn and recognize objects while scanning a scene with eye movements, without causing a combinatorial explosion in the number of cells that are needed? How does the brain avoid the problem of erroneously classifying parts of different objects together at the same or different positions in a visual scene? In monkeys and humans, a key area for such invariant object category learning and recognition is the inferotemporal cortex (IT). A neural model is proposed to explain how spatial and object attention coordinate the ability of IT to learn invariant category representations of objects that are seen at multiple positions, sizes, and viewpoints. The model clarifies how interactions within a hierarchy of processing stages in the visual brain accomplish this. These stages include the retina, lateral geniculate nucleus, and cortical areas V1, V2, V4, and IT in the brain’s What cortical stream, as they interact with spatial attention processes within the parietal cortex of the Where cortical stream. The model builds upon the ARTSCAN model, which proposed how view-invariant object representations are generated. The positional ARTSCAN (pARTSCAN) model proposes how the following additional processes in the What cortical processing stream also enable position-invariant object representations to be learned: IT cells with persistent activity, and a combination of normalizing object category competition and a view-to-object learning law which together ensure that unambiguous views have a larger effect on object recognition than ambiguous views. The model explains how such invariant learning can be fooled when monkeys, or other primates, are presented with an object that is swapped with another object during eye movements to foveate the original object. The swapping procedure is predicted to prevent the reset of spatial attention, which would otherwise keep the representations of multiple objects from being combined by learning. Li and DiCarlo (2008) have presented neurophysiological data from monkeys showing how unsupervised natural experience in a target swapping experiment can rapidly alter object representations in IT. The model quantitatively simulates the swapping data by showing how the swapping procedure fools the spatial attention mechanism. More generally, the model provides a unifying framework, and testable predictions in both monkeys and humans, for understanding object learning data using neurophysiological methods in monkeys, and spatial attention, episodic learning, and memory retrieval data using functional imaging methods in humans.     

Visual experience modulates whole‐brain connectivity dynamics: A resting‐state fMRI study using the model of radiologists

Visual expertise refers to proficiency in visual recognition. It is attributed to accumulated visual experience in a specific domain and manifests in widespread neural activities that extend well beyond the visual cortex to multiple high‐level brain areas. An extensive body of studies has centered on the neural mechanisms underlying a distinctive domain of visual expertise, while few studies elucidated how visual experience modulates resting‐state whole‐brain connectivity dynamics. The current study bridged this gap by modeling the subtle alterations in interregional spontaneous connectivity patterns with a group of superior radiological interns. Functional connectivity analysis was based on functional brain segmentation, which was derived from a data‐driven clustering approach to discriminate subtle changes in connectivity dynamics. Our results showed there was radiographic visual experience accompanied with integration within brain circuits supporting visual processing and decision making, integration across brain circuits supporting high‐order functions, and segregation between high‐order and low‐order brain functions. Also, most of these alterations were significantly correlated with individual nodule identification performance. Our results implied that visual expertise is a controlled, interactive process that develops from reciprocal interactions between the visual system and multiple top‐down factors, including semantic knowledge, top‐down attentional control, and task relevance, which may enhance participants'' local brain functional integration to promote their acquisition of specific visual information and modulate the activity of some regions for lower‐order visual feature processing to filter out nonrelevant visual details. The current findings may provide new ideas for understanding the central mechanism underlying the formation of visual expertise.     

Corticocortical connections among visual areas in the cat

The cortical interconnections of 17 visual areas in the cat were studied by making single injections through recording micropipettes of the neuroanatomical tracers 3H-leucine and horseradish peroxidase (HRP) into the visual cortex of 40 adult animals. Coronal sections from each of the brains were analyzed for location of silver grains and HRP-filled neurons. There are five main results: (1) all corticocortical connections among visual areas are reciprocal. (2) Each cortical visual area has a unique set of cortical connections; the cortical targets of no two cortical visual areas are identical. (3) There is a vast and complicated pattern of connections among the visual areas which implies that there are numerous parallel circuits which run through any one visual area. (4) The connections among the cortical visual areas link retinotopically similar loci and are consistent with the visuotopic maps which microelectrode recording experiments have provided. (5) The connections among visual cortical areas often originate from, or terminate in, discontinuous patches within each area; this result obtains not only for areas 17, 18, 19, and posteromedial lateral suprasylvian area (PMLS), but for at least 13 other areas as well. The data reveal many parallel pathways and suggest multiple functional circuits interconnecting visual cortical areas. Since each visual area has multiple inputs and outputs it may have multiple functions, a different one for each of the circuits of which it is a part.     

Modulation of theta phase synchronization in the human electroencephalogram during a recognition memory task

To the extent that recognition memory relies on interactions among widely distributed neural assemblies across the brain, phase synchronization between brain rhythms may play an important role in meditating those interactions. As the theta rhythm is known to modulate in power during the recognition memory process, we aimed to determine how the phase synchronization of the theta rhythms across the brain changes with recognition memory. Fourteen human participants performed a visual object recognition task in a virtual reality environment. Electroencephalograms were recorded from the scalp of the participants while they either recognized objects that had been presented previously or identified new objects. From the electroencephalogram recordings, we analyzed the phase-locking value of the theta rhythms, which indicates the degree of phase synchronization. We found that the overall phase-locking value recorded during the recognition of previously viewed objects was greater than that recorded during the identification of new objects. Specifically, the theta rhythms became strongly synchronized between the frontal and the left parietal areas during the recognition of previously viewed objects. These results suggest that the recognition memory process may involve an interaction between the frontal and the left parietal cortical regions mediated by theta phase synchronization.     

Feedback inhibition derived from the posterior parietal cortex regulates the neural properties of the mouse visual cortex

Feedback regulation from the higher association areas is thought to control the primary sensory cortex, contribute to the cortical processing of sensory information, and work for higher cognitive functions such as multimodal integration and attentional control. However, little is known about the underlying neural mechanisms. Here, we show that the posterior parietal cortex (PPC) persistently inhibits the activity of the primary visual cortex (V1) in mice. Activation of the PPC causes the suppression of visual responses in V1 and induces the short‐term depression, which is specific to visual stimuli. In contrast, pharmacological inactivation of the PPC or disconnection of cortical pathways from the PPC to V1 results in an effect of transient enhancement of visual responses in V1. Two‐photon calcium imaging demonstrated that the cortical disconnection caused V1 excitatory neurons an enhancement of visual responses and a reduction of orientation selectivity index (OSI). These results show that the PPC regulates the response properties of V1 excitatory neurons. Our findings reveal one of the functions of the PPC, which may contribute to higher brain functions in mice.     

Spatially Specific Working Memory Activity in the Human Superior Colliculus

Theoretically, working memory (WM) representations are encoded by population activity of neurons with distributed tuning across the stored feature. Here, we leverage computational neuroimaging approaches to map the topographic organization of human superior colliculus (SC) and model how population activity in SC encodes WM representations. We first modeled receptive field properties of voxels in SC, deriving a detailed topographic organization resembling that of the primate SC. Neural activity within human (5 male and 1 female) SC persisted throughout a retention interval of several types of modified memory-guided saccade tasks. Assuming an underlying neural architecture of the SC based on its retinotopic organization, we used an encoding model to show that the pattern of activity in human SC represents locations stored in WM. Our tasks and models allowed us to dissociate the locations of visual targets and the motor metrics of memory-guided saccades from the spatial locations stored in WM, thus confirming that human SC represents true WM information. These data have several important implications. They add the SC to a growing number of cortical and subcortical brain areas that form distributed networks supporting WM functions. Moreover, they specify a clear neural mechanism by which topographically organized SC encodes WM representations.SIGNIFICANCE STATEMENT Using computational neuroimaging approaches, we mapped the topographic organization of human superior colliculus (SC) and modeled how population activity in SC encodes working memory (WM) representations, rather than simpler visual or motor properties that have been traditionally associated with the laminar maps in the primate SC. Together, these data both position the human SC into a distributed network of brain areas supporting WM and elucidate the neural mechanisms by which the SC supports WM.     

On the road to invariant recognition: Explaining tradeoff and morph properties of cells in inferotemporal cortex using multiple-scale task-sensitive attentive learning

Visual object recognition is an essential accomplishment of advanced brains. Object recognition needs to be tolerant, or invariant, with respect to changes in object position, size, and view. In monkeys and humans, a key area for recognition is the anterior inferotemporal cortex (ITa). Recent neurophysiological data show that ITa cells with high object selectivity often have low position tolerance. We propose a neural model whose cells learn to simulate this tradeoff, as well as ITa responses to image morphs, while explaining how invariant recognition properties may arise in stages due to processes across multiple cortical areas. These processes include the cortical magnification factor, multiple receptive field sizes, and top-down attentive matching and learning properties that may be tuned by task requirements to attend to either concrete or abstract visual features with different levels of vigilance. The model predicts that data from the tradeoff and image morph tasks emerge from different levels of vigilance in the animals performing them. This result illustrates how different vigilance requirements of a task may change the course of category learning, notably the critical features that are attended and incorporated into learned category prototypes. The model outlines a path for developing an animal model of how defective vigilance control can lead to symptoms of various mental disorders, such as autism and amnesia.     

Dissociable neural correlates of contour completion and contour representation in illusory contour perception

Object recognition occurs even when environmental information is incomplete. Illusory contours (ICs), in which a contour is perceived though the contour edges are incomplete, have been extensively studied as an example of such a visual completion phenomenon. Despite the neural activity in response to ICs in visual cortical areas from low (V1 and V2) to high (LOC: the lateral occipital cortex) levels, the details of the neural processing underlying IC perception are largely not clarified. For example, how do the visual areas function in IC perception and how do they interact to archive the coherent contour perception? IC perception involves the process of completing the local discrete contour edges (contour completion) and the process of representing the global completed contour information (contour representation). Here, functional magnetic resonance imaging was used to dissociate contour completion and contour representation by varying each in opposite directions. The results show that the neural activity was stronger to stimuli with more contour completion than to stimuli with more contour representation in V1 and V2, which was the reverse of that in the LOC. When inspecting the neural activity change across the visual pathway, the activation remained high for the stimuli with more contour completion and increased for the stimuli with more contour representation. These results suggest distinct neural correlates of contour completion and contour representation, and the possible collaboration between the two processes during IC perception, indicating a neural connection between the discrete retinal input and the coherent visual percept. Hum Brain Mapp 33:2407–2414, 2012. © 2011 Wiley Periodicals, Inc.     

A neural circuit basis for spatial working memory.

The maintenance of a mental image in memory over a time scale of seconds is mediated by the persistent discharges of neurons in a distributed brain network. The representation of the spatial location of a remembered visual stimulus has been studied most extensively and provides the best-understood model of how mnemonic information is encoded in the brain. Neural correlates of spatial working memory are manifested in multiple brain areas, including the prefrontal and parietal association cortices. Spatial working memory ability is severely compromised in schizophrenia, a condition that has been linked to prefrontal cortical malfunction. Recent computational modeling work, in interplay with physiological studies of behaving monkeys, has begun to identify microcircuit properties and neural dynamics that are sufficient to generate memory-related persistent activity in a recurrent network of excitatory and inhibitory neurons during spatial working memory. This review summarizes recent results and discusses issues of current debate. It is argued that understanding collective neural dynamics in a recurrent microcircuit provides a key step in bridging the gap between network memory function and its underlying cellular mechanisms. Progress in this direction will shed fundamental insights into the neural basis of spatial working memory impairment associated with mental disorders.     

The relationship between visual perception and visual mental imagery: a reappraisal of the neuropsychological evidence

Visual perception and visual mental imagery, the faculty whereby we can revisualise a visual item from memory, have often been regarded as cognitive functions subserved by common mechanisms. Thus, the leading cognitive model of visual mental imagery holds that visual perception and visual imagery share a number of mental operations, and rely upon common neural structures, including early visual cortices. In particular, a single visual buffer would be used "bottom-up" to display visual percepts and "top-down" to display internally generated images. The proposed neural substrate for this buffer consists of some cortical visual areas organised retinotopically, that is, the striate and extrastriate occipital areas. Empirical support for this model came from the report of brain-damaged patients showing an imagery deficit which parallels a perceptual impairment in the same cognitive domain. However, recent reports of patients showing double dissociations between perception and imagery abilities challenged the perception-imagery equivalence hypothesis from the functional point of view. From the anatomical point of view, the available evidence suggests that occipital damage is neither necessary nor sufficient to produce imagery deficits. On the other hand, extensive left temporal damage often accompanies imagery deficits for object form or colour. Thus, visual mental imagery abilities might require the integrity of brain areas related to vision, but at an higher level of integration than previously proposed.     

Organization of face and object recognition in modular neural network models

There is strong evidence that face processing in the brain is localized. The double dissociation between prosopagnosia, a face recognition deficit occurring after brain damage, and visual object agnosia, difficulty recognizing other kinds of complex objects, indicates that face and non-face object recognition may be served by partially independent neural mechanisms. In this paper, we use computational models to show how the face processing specialization apparently underlying prosopagnosia and visual object agnosia could be attributed to (1) a relatively simple competitive selection mechanism that, during development, devotes neural resources to the tasks they are best at performing, (2) the developing infant's need to perform subordinate classification (identification) of faces early on, and (3) the infant's low visual acuity at birth. Inspired by de Schonen, Mancini and Liegeois' arguments (1998) [de Schonen, S., Mancini, J., Liegeois, F. (1998). About functional cortical specialization: the development of face recognition. In: F. Simon & G. Butterworth, The development of sensory, motor, and cognitive capacities in early infancy (pp. 103–116). Hove, UK: Psychology Press] that factors like these could bias the visual system to develop a processing subsystem particularly useful for face recognition, and Jacobs and Kosslyn's experiments (1994) [Jacobs, R. A., & Kosslyn, S. M. (1994). Encoding shape and spatial relations—the role of receptive field size in coordination complementary representations. Cognitive Science, 18(3), 361–368] in the mixtures of experts (ME) modeling paradigm, we provide a preliminary computational demonstration of how this theory accounts for the double dissociation between face and object processing. We present two feed-forward computational models of visual processing. In both models, the selection mechanism is a gating network that mediates a competition between modules attempting to classify input stimuli. In Model I, when the modules are simple unbiased classifiers, the competition is sufficient to achieve enough of a specialization that damaging one module impairs the model's face recognition more than its object recognition, and damaging the other module impairs the model's object recognition more than its face recognition. However, the model is not completely satisfactory because it requires a search of parameter space. With Model II, we explore biases that lead to more consistent specialization. We bias the modules by providing one with low spatial frequency information and the other with high spatial frequency information. In this case, when the model's task is subordinate classification of faces and superordinate classification of objects, the low spatial frequency network shows an even stronger specialization for faces. No other combination of tasks and inputs shows this strong specialization. We take these results as support for the idea that something resembling a face processing “module” could arise as a natural consequence of the infant's developmental environment without being innately specified.