Top-down Connections Research Articles

Predicting effective connectivity from resting‐state networks in healthy elderly and patients with prodromal Alzheimer's disease

Using functional neuroimaging techniques two aspects of functional integration in the human brain have been investigated, functional connectivity and effective connectivity. In this study we examined both connectivity types in parallel within an executive attention network during rest and while performing an attention task. We analyzed the predictive value of resting-state functional connectivity on task-induced effective connectivity in patients with prodromal Alzheimer's disease (AD) and healthy elderly. We found that in healthy elderly, functional connectivity was a significant predictor for effective connectivity, however, it was frequency-specific. Effective top-down connectivity emerging from prefrontal areas was related with higher frequencies of functional connectivity (e.g., 0.08-0.15 Hz), in contrast to effective bottom-up connectivity going to prefrontal areas, which was related to lower frequencies of functional connectivity (e.g., 0.001-0.03 Hz). In patients, the prediction of effective connectivity by functional connectivity was disturbed. We conclude that functional connectivity and effective connectivity are interrelated in healthy brains but this relationship is aberrant in very early AD.

Abnormalities in fronto-striatal connectivity within language networks relate to differences in grey-matter heterogeneity in Asperger syndrome

Asperger syndrome (AS) is an Autism Spectrum Disorder (ASD) characterised by qualitative impairment in the development of emotional and social skills with relative preservation of general intellectual abilities, including verbal language. People with AS may nevertheless show atypical language, including rate and frequency of speech production. We previously observed that abnormalities in grey matter homogeneity (measured with texture analysis of structural MR images) in AS individuals when compared with controls are also correlated with the volume of caudate nucleus. Here, we tested a prediction that these distributed abnormalities in grey matter compromise the functional integrity of brain networks supporting verbal communication skills. We therefore measured the functional connectivity between caudate nucleus and cortex during a functional neuroimaging study of language generation (verbal fluency), applying psycho-physiological interaction (PPI) methods to test specifically for differences attributable to grey matter heterogeneity in AS participants. Furthermore, we used dynamic causal modelling (DCM) to characterise the causal directionality of these differences in interregional connectivity during word production. Our results revealed a diagnosis-dependent influence of grey matter heterogeneity on the functional connectivity of the caudate nuclei with right insula/inferior frontal gyrus and anterior cingulate, respectively with the left superior frontal gyrus and right precuneus. Moreover, causal modelling of interactions between inferior frontal gyri, caudate and precuneus, revealed a reliance on bottom-up (stimulus-driven) connections in AS participants that contrasted with a dominance of top-down (cognitive control) connections from prefrontal cortex observed in control participants. These results provide detailed support for previously hypothesised central disconnectivity in ASD and specify discrete brain network targets for diagnosis and therapy in ASD.

Artificial vision by multi-layered neural networks: Neocognitron and its advances

The neocognitron is a neural network model proposed by Fukushima (1980). Its architecture was suggested by neurophysiological findings on the visual systems of mammals. It is a hierarchical multi-layered network. It acquires the ability to robustly recognize visual patterns through learning. Although the neocognitron has a long history, modifications of the network to improve its performance are still going on. For example, a recent neocognitron uses a new learning rule, named add-if-silent, which makes the learning process much simpler and more stable. Nevertheless, a high recognition rate can be kept with a smaller scale of the network. Referring to the history of the neocognitron, this paper discusses recent advances in the neocognitron. We also show that various new functions can be realized by, for example, introducing top-down connections to the neocognitron: mechanism of selective attention, recognition and completion of partly occluded patterns, restoring occluded contours, and so on.

Deconstructing the Architecture of Dorsal and Ventral Attention Systems with Dynamic Causal Modeling

Attentional orientation to a spatial cue and reorientation-after invalid cueing-are mediated by two distinct networks in the human brain. A bilateral dorsal frontoparietal network, comprising the intraparietal sulcus (IPS) and the frontal eye fields (FEF), controls the voluntary deployment of attention and may modulate visual cortex in preparation for upcoming stimulation. In contrast, reorienting attention to invalidly cued targets engages a right-lateralized ventral frontoparietal network comprising the temporoparietal junction (TPJ) and ventral frontal cortex. The present fMRI study investigated the functional architecture of these two attentional systems by characterizing effective connectivity during lateralized orienting and reorienting of attention, respectively. Subjects performed a modified version of Posner's location-cueing paradigm. Dynamic causal modeling (DCM) of regional responses in the dorsal and ventral network, identified in a conventional (SPM) whole-brain analysis, was used to compare different functional architectures. Bayesian model selection showed that top-down connections from left and right IPS to left and right visual cortex, respectively, were modulated by the direction of attention. Moreover, model evidence was highest for a model with directed influences from bilateral IPS to FEF, and reciprocal coupling between right and left FEF. Invalid cueing enhanced forward connections from visual areas to right TPJ, and directed influences from right TPJ to right IPS and IFG (inferior frontal gyrus). These findings shed further light on the functional organization of the dorsal and ventral attentional network and support a context-sensitive lateralization in the top-down (backward) mediation of attentional orienting and the bottom-up (forward) effects of invalid cueing.

Depression-Biased Reverse Plasticity Rule Is Required for Stable Learning at Top-Down Connections

Top-down synapses are ubiquitous throughout neocortex and play a central role in cognition, yet little is known about their development and specificity. During sensory experience, lower neocortical areas are activated before higher ones, causing top-down synapses to experience a preponderance of post-synaptic activity preceding pre-synaptic activity. This timing pattern is the opposite of that experienced by bottom-up synapses, which suggests that different versions of spike-timing dependent synaptic plasticity (STDP) rules may be required at top-down synapses. We consider a two-layer neural network model and investigate which STDP rules can lead to a distribution of top-down synaptic weights that is stable, diverse and avoids strong loops. We introduce a temporally reversed rule (rSTDP) where top-down synapses are potentiated if post-synaptic activity precedes pre-synaptic activity. Combining analytical work and integrate-and-fire simulations, we show that only depression-biased rSTDP (and not classical STDP) produces stable and diverse top-down weights. The conclusions did not change upon addition of homeostatic mechanisms, multiplicative STDP rules or weak external input to the top neurons. Our prediction for rSTDP at top-down synapses, which are distally located, is supported by recent neurophysiological evidence showing the existence of temporally reversed STDP in synapses that are distal to the post-synaptic cell body.

Brain Circuits of Methamphetamine Place Reinforcement Learning: The Role of the Hippocampus‐VTA Loop

The reinforcing effects of addictive drugs including methamphetamine (METH) involve the midbrain ventral tegmental area (VTA). VTA is primary source of dopamine (DA) to the nucleus accumbens (NAc) and the ventral hippocampus (VHC). These three brain regions are functionally connected through the hippocampal-VTA loop that includes two main neural pathways: the bottom-up pathway and the top-down pathway. In this paper, we take the view that addiction is a learning process. Therefore, we tested the involvement of the hippocampus in reinforcement learning by studying conditioned place preference (CPP) learning by sequentially conditioning each of the three nuclei in either the bottom-up order of conditioning; VTA, then VHC, finally NAc, or the top-down order; VHC, then VTA, finally NAc. Following habituation, the rats underwent experimental modules consisting of two conditioning trials each followed by immediate testing (test 1 and test 2) and two additional tests 24 h (test 3) and/or 1 week following conditioning (test 4). The module was repeated three times for each nucleus. The results showed that METH, but not Ringer's, produced positive CPP following conditioning each brain area in the bottom-up order. In the top-down order, METH, but not Ringer's, produced either an aversive CPP or no learning effect following conditioning each nucleus of interest. In addition, METH place aversion was antagonized by coadministration of the N-methyl-d-aspartate (NMDA) receptor antagonist MK801, suggesting that the aversion learning was an NMDA receptor activation-dependent process. We conclude that the hippocampus is a critical structure in the reward circuit and hence suggest that the development of target-specific therapeutics for the control of addiction emphasizes on the hippocampus-VTA top-down connection.

Once upon a time there was complex numerical estimation

Once upon a time there was complex numerical estimation

Differences in Early Dynamic Connectivity between Visual Expansion and Contraction Stimulations Revealed by an fMRI‐Directed MEG Approach

The human visual system responds asymmetrically to visual motion stimuli in opposite directions due to the involvement of the same brain areas but different operating processes. The expansion mode is thought to invoke a vigilance mechanism, whereas the contraction mode does not. To investigate discrepancies between these modes, we produced dynamic connectivity maps based on mutual information between visual-evoked dipole sources of magnetoencephalography, which were steered by visual activity patterns in functional magnetic resonance imaging under two motion-stimulus modes. In the expansion mode, information was conveyed from V1 at 50-75 ms after motion onset, fed forward to V3A and then V5. Top-down connectivity paths were evident after a latency of 100 ms. Many of these interactions occurred within 200 ms. However, in the contraction mode, information was conveyed from V3A to V5, followed by feedback, but regained from V1 after a latency of 250 ms. Although these interactions were delayed by about 250 ms, they were completed within 500 ms. These findings show that detect spatiotemporal differences between expansion and contraction modes can be readily detected using time-flow charts. Moreover, delay interactions could be insensitive to object motion away from the observer.

Incremental Online Object Learning in a Vehicular Radar-Vision Fusion Framework

In this paper, we propose an object learning system that incorporates sensory information from an automotive radar system and a video camera. The radar system provides coarse attention for the focus of visual analysis on relatively small areas within the image plane. The attended visual areas are coded and learned by a three-layer neural network utilizing what is called in-place learning: Each neuron is responsible for the learning of its own processing characteristics within the connected network environment, through inhibitory and excitatory connections with other neurons. The modeled bottom-up, lateral, and top-down connections in the network enable sensory sparse coding, unsupervised learning, and supervised learning to occur concurrently. This paper is applied to learn two types of encountered objects in multiple outdoor driving settings. Cross-validation results show that the overall recognition accuracy is above 95% for the radar-attended window images. In comparison with the uncoded representation and purely unsupervised learning (without top-down connection), the proposed network improves the overall recognition rate by 15.93% and 6.35%, respectively. The proposed system is also compared favorably with other learning algorithms. The result indicates that our learning system is the only one that is fit for incremental and online object learning in a real-time driving environment.

Impaired Hierarchical Control Within the Lateral Prefrontal Cortex in Schizophrenia

In schizophrenia, disturbances of cognitive control have been associated with impaired functional specialization within the lateral prefrontal cortex (LPFC), but little is known about the functional interactions between specialized LPFC subregions. Here, we addressed this question with a recent model that describes the LPFC functioning as a cascade of control processes along a rostrocaudal axis, whereby anterior frontal regions influence the processing in posterior frontal regions to guide action selection on the basis of the temporal structure of information. We assessed effective connectivity within the rostrocaudal axis of the LPFC by means of functional magnetic resonance imaging in 15 schizophrenic patients and 14 matched healthy control subjects with structural equation modeling and psychophysiological interactions. In healthy subjects, activity in the left caudal LPFC regions was under the influence of left rostral LPFC regions when controlling information conveyed by past events. By contrast, schizophrenic patients failed to demonstrate significant effective connectivity from rostral to caudal LPFC regions in both hemispheres. The hierarchical control along the rostrocaudal axis of the LPFC is impaired in schizophrenia. This provides the first evidence of a top-down functional disconnection within the LPFC in this disorder. This disruption of top-down connectivity from rostral to caudal LPFC regions observed in patients might affect their ability to select the appropriate sets of stimulus-response associations in the caudal LPFC on the basis of information conveyed by past events. This impaired hierarchical control within the LPFC could result from poorly encoded contextual information due to abnormal computations in the caudal LPFC.

Off-line memory reprocessing in a recurrent neuronal network formed by unsupervised learning

AbstractIn the visual cortex, memory traces for complex objects are embedded into a scaffold of feed-forward and recurrent connectivity of the hierarchically organized visual pathway. Strong evidence suggests that consolidation of the memory traces in such a memory network depends on an off-line reprocessing done in the sleep state or during restful waking. It remains largely unclear, what plasticity mechanisms are involved in this consolidation process and what changes are induced in the network during memory reprocessing in the off-line regime. Here we focus on the functional consequences off-line reprocessing has in a hierarchical recurrent neuronal network that learns different person identities from natural face images in an unsupervised manner. Due to the inherently self-exciting, but competitive unit dynamics, the two-layered network is able to self-generate sparse activity even in the absence of external input in an off-line regime. In this regime, the network replays the memory content established during preceding on-line learning. Remarkably, this off-line memory replay turns out to be highly beneficial for the network recognition performance. The benefit is articulated after the off-line regime in a strong boost of identity recognition rate on the alternative face views to which the network has not been exposed during learning. Performance of both network layers is affected by the boost. Surprisingly, the positive effect is independent of synapse-specific plasticity, relying completely on a synapse-unspecific mechanism of homeostatic activity regulation that tunes network unit excitability. Comparing further a purely feed-forward configuration of the network with its fully recurrent original version reveals a stronger boost in recognition performance for the latter after the off-line reprocessing. These findings suggest that the off-line memory reprocessing enhances generalization capability of the hierarchical recurrent network by improving communication of contextual cues mediated via recurrent lateral and top-down connectivity.

An examination of orthographic and phonological processing using the task-choice procedure

The task-choice procedure provides a way for assessing whether stimuli are processed immediately upon presentation and in parallel with other cognitive operations. In this procedure, the task changes on a trial-by-trial basis and the cue informing participants about the task appears either before or simultaneously with the target, which is either degraded or clear. Of interest is whether the effect of stimulus clarity will disappear when the cue is presented simultaneously with the target, suggesting capacity-free processing, or whether the effect of stimulus clarity will remain, suggesting target processing is delayed. Besner and Care developed this procedure using nonword targets and found that phonological information was not extracted in parallel with deciphering the task cue. The current experiment examined whether phonological and orthographic information could be extracted from word and nonword stimuli in a capacity-free manner. Results indicate that in both tasks some processing does occur in a capacity-free manner when words are used but not when nonwords are used. These data may be consistent with interactive activation models which posit top-down lexical connections that facilitate the extraction of sublexical codes.

Acquisition of nonlinear forward optics in generative models: Two-stage “downside-up” learning for occluded vision

We propose a two-stage learning method which implements occluded visual scene analysis into a generative model, a type of hierarchical neural network with bi-directional synaptic connections. Here, top-down connections simulate forward optics to generate predictions for sensory driven low-level representation, whereas bottom-up connections function to send the prediction error, the difference between the sensory based and the predicted low-level representation, to higher areas. The prediction error is then used to update the high-level representation to obtain better agreement with the visual scene. Although the actual forward optics is highly nonlinear and the accuracy of simulated forward optics is crucial for these types of models, the majority of previous studies have only investigated linear and simplified cases of forward optics. Here we take occluded vision as an example of nonlinear forward optics, where an object in front completely masks out the object behind. We propose a two-staged learning method inspired by the staged development of infant visual capacity. In the primary learning stage, a minimal set of object basis is acquired within a linear generative model using the conventional unsupervised learning scheme. In the secondary learning stage, an auxiliary multi-layer neural network is trained to acquire nonlinear forward optics by supervised learning. The important point is that the high-level representation of the linear generative model serves as the input and the sensory driven low-level representation provides the desired output. Numerical simulations show that occluded visual scene analysis can indeed be implemented by the proposed method. Furthermore, considering the format of input to the multi-layer network and analysis of hidden-layer units leads to the prediction that whole object representation of partially occluded objects, together with complex intermediate representation as a consequence of nonlinear transformation from non-occluded to occluded representation may exist in the low-level visual system of the brain.

Abnormally Reduced Dorsomedial Prefrontal Cortical Activity and Effective Connectivity With Amygdala in Response to Negative Emotional Faces in Postpartum Depression

Postpartum major depression is a significant public health problem that strikes 15% of new mothers and confers adverse consequences for mothers, children, and families. The neural mechanisms involved in postpartum depression remain unknown, but brain processing of affective stimuli appears to be involved in other affective disorders. The authors examined activity in response to negative emotional faces in the dorsomedial pre-frontal cortex and amygdala, key emotion regulatory neural regions of importance to both mothering and depression. Postpartum healthy mothers (N=16) and unmedicated depressed mothers (N=14) underwent functional magnetic resonance imaging blood-oxygen-level-dependent acquisition during a block-designed face versus shape matching task. A two-way analysis of variance was performed examining main effects of condition and group and group-by-condition interaction on activity in bilateral dorsomedial prefrontal cortical and amygdala regions of interest. Depressed mothers relative to healthy mothers had significantly reduced left dorsomedial prefrontal cortical face-related activity. In depressed mothers, there was also a significant negative correlation between left amygdala activity and postpartum depression severity and a significant positive correlation between right amygdala activity and absence of infant-related hostility. There was reliable top-down connectivity from the left dorsomedial prefrontal cortex to the left amygdala in healthy, but not depressed, mothers. Significantly diminished dorsomedial prefrontal cortex activity and dorsomedial prefrontal cortical-amygdala effective connectivity in response to negative emotional faces may represent an important neural mechanism, or effect, of postpartum depression. Reduced amygdala activity in response to negative emotional faces is associated with greater postpartum depression severity and more impaired maternal attachment processes in postpartum depressed mothers.

Discovering binary codes for documents by learning deep generative models.

We describe a deep generative model in which the lowest layer represents the word-count vector of a document and the top layer represents a learned binary code for that document. The top two layers of the generative model form an undirected associative memory and the remaining layers form a belief net with directed, top-down connections. We present efficient learning and inference procedures for this type of generative model and show that it allows more accurate and much faster retrieval than latent semantic analysis. By using our method as a filter for a much slower method called TF-IDF we achieve higher accuracy than TF-IDF alone and save several orders of magnitude in retrieval time. By using short binary codes as addresses, we can perform retrieval on very large document sets in a time that is independent of the size of the document set using only one word of memory to describe each document.

Top-down connections as abstract and temporal context for self-organization and categorization

There are many bottom-up connections from inferior temporal cortex (ITC) to prefrontal cortex (PFC) and many top-down connections from PFC to ITC. It is hypothesized that neurons in PFC perform behaviorally-relevant category binding, while neurons in ITC are responsive to high-level visual features [1]. Therefore it is thought the bottom-up connections provide information about detected high-level features to PFC, which binds them together for categorization. But all the computational roles of the top-down connections are currently not known, specifically in development and learning. We seek to verify two general ideas about top-down connections via simulation: that they could act as the impetus of category-specific self-organization, e.g., seen in the fusiform face area (FFA) and parahippocampal place area (PPA), and that they can act as a “bias” (memory store) for biasing the ITC features [2]. We built networks with three interconnected neuronal layers: a sensory area (layer one), a feature representation area (layer-two: ITC), and a category-behavior area (layer-three: PFC). Each layer-two neuron receives excitatory inputs from the bottom-up, laterally, and top-down and each layer-three neuron has bottom-up inputs. Neurons compete with others on the same layer through lateral inhibition. Neurons that are not firing-inhibited learn through a Hebbian learning algorithm [3], in which the strength of synaptic learning is based on presynaptic and postsynaptic potentials. We used visual stimuli (images) of 26 classes of objects that sequentially rotate in depth, presenting them randomly, and learning in a supervised fashion. We observed two results: developed ITC neurons became more specialized to represent a particular class and grouped class areas emerged on the ITC neuronal plane. In contrast to slow feature analysis methods [4], our network’s development does not depend on slowly changing inputs, but instead the correlations between the category information from PFC and the true class of the stimulus. This might explain how PPA, which represents “places” --- a very abstract category containing members that are experienced at different times --- could develop. Networks with top-down connections showed an average error rate reduction of 63% over networks where top-down connections were disabled. The classification rate of a 40 x 40 layer-two was 95% recognition when operating in feedforward. When also utilizing top-down connections, performance reached nearly 100%, but errors were introduced during the periods when an object transitioned to another (yet the network successfully recovered). This is consistent with the idea that top-down connections can bias lower-level features based on currently detected categories.

Where-what networks for motor invariance without any internal master map

The adult brain appears to have a capability of location invariance: No matter where on the retina an object appears, the brain recognizes the object. Yet, this does not mean that the brain drops the location information, since it needs such information for arm reaching. Mishkin and coworkers 1983 [1] reported that the dorsal and ventral streams of the brain are correlated to space (“where”) and object (“what”) information, respectively, based largely on their brain lesion studies. Many later experimental studies verified and enriched this discovery, the working and learning of these two streams have been elusive (Deco & Rolls 2004 [2]). It has been known that feedback connections are widely present along these streams, but computational understanding and analysis are lacking. On the other hand, the sensory cortex alone seems to use distributed representations. Each feature neuron has a receptive field, corresponding to a patch in the retina. There are multiple nearby neurons whose receptive fields almost completely overlap and they detect different features of the overlapping patches (e.g., each for a different edge orientation). However, such distributed “patch representations” must be combined somehow to give rise to behaviors that demonstrate invariant object recognition. Ann Treisman [2] and David Van Essen et al. [4] proposed the existence of a mater feature map. Following neuro-anatomical data, our visuomotor model Where What Network (WWN) suggests that to understand the causality of the above phenomenon, it is beneficial to go beyond PP and IT to include the premotor and motor areas in the frontal cortex. We introduce two motor areas as the integral parts of cortical object representation: location motor (LM) and type motor (TM). The former correlates the frontal eye field (FEF) and the location-relevant control areas in the pre-motor and motor areas. The latter corresponds to the ventral frontal cortex (VFC) and the verbal control area in the pre-motor and motor areas. The dorsal stream plus LM learns type invariance and location specificity (for, e.g., arm reaching). The ventral stream plus TM learns location invariance and type specificity (for, e.g., pronounce the object type). Bottom-up and top-down connections from LM and TM dynamically wire connections and shape the corresponding streams, resulting in complementary representations: invariance with one is specificity with the other. WWNs were tested to deal with the tightly intertwined attention and recognition for vision in the presence of complex backgrounds. Each of attention and recognition has been modeled separately in previous work, e.g., visual saliency guides covert attention sifts. How the visual cortex deals with both attention and recognition from complex natural backgrounds conjunctively has been elusive. WWN gives the first biological plausible theory for this joint problem. With general object in complex new backgrounds, WWN reported 95% in classification rate and under 2-pixel location error, when about 75% images areas are from unknown complex backgrounds. Each WWN epigenetically generates and adapts emergent representations using Hebbian like neuronal learning mechanisms. WWN explains how top-down attention originates from LM for location-based and TM for type-based top-down attention. This model does not need an the appearance-kept internal master feature maps proposed earlier.

Neural mass models for mimicking brain signals – impact of extrinsic inputs on interneurons and dendritic time constants

Introduction In order to reproduce brain signals like electroencephalography (EEG), one has to consider biological inspired and plausible models like neural mass models. The neuronal currents underlying the generation of EEG are assumed to be generated by the postsynaptic potentials (PSPs) of pyramidal cells. A cortical area can be described by three neural masses: pyramidal cells, excitatory and inhibitory interneurons, strongly interacting by positive and negative feedback loops [1]. Due to the cortical organization, shortand long-range connections establish afferent pathways on interneurons, which are of major importance for top-down and lateral connections [2]. To date, there is no systematic analysis of the impact of extrinsic inputs on pyramidal cells and interneurons.

Abnormal Amygdala-Prefrontal Effective Connectivity to Happy Faces Differentiates Bipolar from Major Depression

Bipolar disorder is frequently misdiagnosed as major depressive disorder, delaying appropriate treatment and worsening outcome for many bipolar individuals. Emotion dysregulation is a core feature of bipolar disorder. Measures of dysfunction in neural systems supporting emotion regulation might therefore help discriminate bipolar from major depressive disorder. Thirty-one depressed individuals-15 bipolar depressed (BD) and 16 major depressed (MDD), DSM-IV diagnostic criteria, ages 18-55 years, matched for age, age of illness onset, illness duration, and depression severity-and 16 age- and gender-matched healthy control subjects performed two event-related paradigms: labeling the emotional intensity of happy and sad faces, respectively. We employed dynamic causal modeling to examine significant among-group alterations in effective connectivity (EC) between right- and left-sided neural regions supporting emotion regulation: amygdala and orbitomedial prefrontal cortex (OMPFC). During classification of happy faces, we found profound and asymmetrical differences in EC between the OMPFC and amygdala. Left-sided differences involved top-down connections and discriminated between depressed and control subjects. Furthermore, greater medication load was associated with an amelioration of this abnormal top-down EC. Conversely, on the right side the abnormality was in bottom-up EC that was specific to bipolar disorder. These effects replicated when we considered only female subjects. Abnormal, left-sided, top-down OMPFC-amygdala and right-sided, bottom-up, amygdala-OMPFC EC during happy labeling distinguish BD and MDD, suggesting different pathophysiological mechanisms associated with the two types of depression.

Dynamic Causal Modeling of the Response to Frequency Deviants

This article describes the use of dynamic causal modeling to test hypotheses about the genesis of evoked responses. Specifically, we consider the mismatch negativity (MMN), a well-characterized response to deviant sounds and one of the most widely studied evoked responses. There have been several mechanistic accounts of how the MMN might arise. It has been suggested that the MMN results from a comparison between sensory input and a memory trace of previous input, although others have argued that local adaptation, due to stimulus repetition, is sufficient to explain the MMN. Thus the precise mechanisms underlying the generation of the MMN remain unclear. This study tests some biologically plausible spatiotemporal dipole models that rest on changes in extrinsic top-down connections (that enable comparison) and intrinsic changes (that model adaptation). Dynamic causal modeling suggested that responses to deviants are best explained by changes in effective connectivity both within and between cortical sources in a hierarchical network of distributed sources. Our model comparison suggests that both adaptation and memory comparison operate in concert to produce the early (N1 enhancement) and late (MMN) parts of the response to frequency deviants. We consider these mechanisms in the light of predictive coding and hierarchical inference in the brain.