Computing with a canonical neural circuits model with pool normalization and modulating feedback

Canonical circuit computations for computer vision

Advanced computer vision mechanisms have been inspired by neuroscientific findings. However, with the focus on improving benchmark achievements, technical solutions have been shaped by application and engineering constraints. This includes the training of neural networks which led to the development of feature detectors optimally suited to the application domain. However, the limitations of such approaches motivate the need to identify computational principles, or motifs, in biological vision that can enable further foundational advances in machine vision. We propose to utilize structural and functional principles of neural systems that have been largely overlooked. They potentially provide new inspirations for computer vision mechanisms and models. Recurrent feedforward, lateral, and feedback interactions characterize general principles underlying processing in mammals. We derive a formal specification of core computational motifs that utilize these principles. These are combined to define model mechanisms for visual shape and motion processing. We demonstrate how such a framework can be adopted to run on neuromorphic brain-inspired hardware platforms and can be extended to automatically adapt to environment statistics. We argue that the identified principles and their formalization inspires sophisticated computational mechanisms with improved explanatory scope. These and other elaborated, biologically inspired models can be employed to design computer vision solutions for different tasks and they can be used to advance neural network architectures of learning.

Listen to the Brain-Auditory Sound Source Localization in Neuromorphic Computing Architectures

Conventional processing of sensory input often relies on uniform sampling leading to redundant information and unnecessary resource consumption throughout the entire processing pipeline. Neuromorphic computing challenges these conventions by mimicking biology and employing distributed event-based hardware. Based on the task of lateral auditory sound source localization (SSL), we propose a generic approach to map biologically inspired neural networks to neuromorphic hardware. First, we model the neural mechanisms of SSL based on the interaural level difference (ILD). Afterward, we identify generic computational motifs within the model and transform them into spike-based components. A hardware-specific step then implements them on neuromorphic hardware. We exemplify our approach by mapping the neural SSL model onto two platforms, namely the IBM TrueNorth Neurosynaptic System and SpiNNaker. Both implementations have been tested on synthetic and real-world data in terms of neural tunings and readout characteristics. For synthetic stimuli, both implementations provide a perfect readout (100% accuracy). Preliminary real-world experiments yield accuracies of 78% (TrueNorth) and 13% (SpiNNaker), RMSEs of 41∘ and 39∘, and MAEs of 18∘ and 29∘, respectively. Overall, the proposed mapping approach allows for the successful implementation of the same SSL model on two different neuromorphic architectures paving the way toward more hardware-independent neural SSL.

A multi-scale neurodynamic implementation of incremental grouping

Incremental grouping is a process entailing serial binding of distal image elements into a unified object representation. At the neural level, incremental grouping involves propagation of the enhanced firing rate among feature-tuned neurons in the early visual cortex. Here, we developed a multi-resolution neural model of incremental grouping. In the model, propagation of the enhanced firing rate is achieved by computing the activity difference between two sets of units: attentional or A-units, whose firing rate is modulated by their horizontal collaterals, and non-attentional or N-units that receive only feedforward input. The activity difference is computed on dendrites that act as independent computational subunits. The proposed model employs multiple spatial scales to account for a variable speed of incremental grouping. In addition, the model incorporates the L-junction detection network that enables incremental grouping over L-junctions. Computer simulations show that the timing of attentional modulations in the model is comparable with neurophysiological measurements in monkey primary visual cortex.

Divisive normalization unifies disparate response signatures throughout the human visual hierarchy

A canonical neural computation is a mathematical operation applied by the brain in a wide variety of contexts and capable of explaining and unifying seemingly unrelated neural and perceptual phenomena. Here, we use a combination of state-of-the-art experiments (ultra-high-field functional MRI) and mathematical methods (population receptive field [pRF] modeling) to uniquely demonstrate the role of divisive normalization (DN) as the canonical neural computation underlying visuospatial responses throughout the human visual hierarchy. The DN pRF model provides a tool to investigate and interpret the computational processes underlying neural responses in human and animal recordings, but also in clinical and cognitive dimensions.

Neural processing is hypothesized to apply the same mathematical operations in a variety of contexts, implementing so-called canonical neural computations. Divisive normalization (DN) is considered a prime candidate for a canonical computation. Here, we propose a population receptive field (pRF) model based on DN and evaluate it using ultra-high-field functional MRI (fMRI). The DN model parsimoniously captures seemingly disparate response signatures with a single computation, superseding existing pRF models in both performance and biological plausibility. We observe systematic variations in specific DN model parameters across the visual hierarchy and show how they relate to differences in response modulation and visuospatial information integration. The DN model delivers a unifying framework for visuospatial responses throughout the human visual hierarchy and provides insights into its underlying information-encoding computations. These findings extend the role of DN as a canonical computation to neuronal populations throughout the human visual hierarchy.

Dynamic divisive normalization circuits explain and predict change detection in monkey area MT

Sudden changes in visual scenes often indicate important events for behavior. For their quick and reliable detection, the brain must be capable to process these changes as independently as possible from its current activation state. In motion-selective area MT, neurons respond to instantaneous speed changes with pronounced transients, often far exceeding the expected response as derived from their speed tuning profile. We here show that this complex, non-linear behavior emerges from the combined temporal dynamics of excitation and divisive inhibition, and provide a comprehensive mathematical analysis. A central prediction derived from this investigation is that attention increases the steepness of the transient response irrespective of the activation state prior to a stimulus change, and irrespective of the sign of the change (i.e. irrespective of whether the stimulus is accelerating or decelerating). Extracellular recordings of attention-dependent representation of both speed increments and decrements confirmed this prediction and suggest that improved change detection derives from basic computations in a canonical cortical circuitry.

A recurrent circuit implements normalization, simulating the dynamics of V1 activity

The normalization model has been applied to explain neural activity in diverse neural systems including primary visual cortex (V1). The model's defining characteristic is that the response of each neuron is divided by a factor that includes a weighted sum of activity of a pool of neurons. Despite the success of the normalization model, there are three unresolved issues. 1) Experimental evidence supports the hypothesis that normalization in V1 operates via recurrent amplification, i.e., amplifying weak inputs more than strong inputs. It is unknown how normalization arises from recurrent amplification. 2) Experiments have demonstrated that normalization is weighted such that each weight specifies how one neuron contributes to another's normalization pool. It is unknown how weighted normalization arises from a recurrent circuit. 3) Neural activity in V1 exhibits complex dynamics, including gamma oscillations, linked to normalization. It is unknown how these dynamics emerge from normalization. Here, a family of recurrent circuit models is reported, each of which comprises coupled neural integrators to implement normalization via recurrent amplification with arbitrary normalization weights, some of which can recapitulate key experimental observations of the dynamics of neural activity in V1.

Computational principles of neural adaptation for binaural signal integration

Adaptation to statistics of sensory inputs is an essential ability of neural systems and extends their effective operational range. Having a broad operational range facilitates to react to sensory inputs of different granularities, thus is a crucial factor for survival. The computation of auditory cues for spatial localization of sound sources, particularly the interaural level difference (ILD), has long been considered as a static process. Novel findings suggest that this process of ipsi- and contra-lateral signal integration is highly adaptive and depends strongly on recent stimulus statistics. Here, adaptation aids the encoding of auditory perceptual space of various granularities. To investigate the mechanism of auditory adaptation in binaural signal integration in detail, we developed a neural model architecture for simulating functions of lateral superior olive (LSO) and medial nucleus of the trapezoid body (MNTB) composed of single compartment conductance-based neurons. Neurons in the MNTB serve as an intermediate relay population. Their signal is integrated by the LSO population on a circuit level to represent excitatory and inhibitory interactions of input signals. The circuit incorporates an adaptation mechanism operating at the synaptic level based on local inhibitory feedback signals. The model’s predictive power is demonstrated in various simulations replicating physiological data. Incorporating the innovative adaptation mechanism facilitates a shift in neural responses towards the most effective stimulus range based on recent stimulus history. The model demonstrates that a single LSO neuron quickly adapts to these stimulus statistics and, thus, can encode an extended range of ILDs in the ipsilateral hemisphere. Most significantly, we provide a unique measurement of the adaptation efficacy of LSO neurons. Prerequisite of normal function is an accurate interaction of inhibitory and excitatory signals, a precise encoding of time and a well-tuned local feedback circuit. We suggest that the mechanisms of temporal competitive-cooperative interaction and the local feedback mechanism jointly sensitize the circuit to enable a response shift towards contra-lateral and ipsi-lateral stimuli, respectively.

Why are we more precise in localizing a sound after hearing it several times? Adaptation to the statistics of a stimulus plays a crucial role in this. The present article investigates the abilities of a neural adaptation mechanism for improved localization skills based on a neural network model. Adaptation to stimulus statistics is very prominent in sensory systems of animals and allows them to respond to a wide range of stimuli, thus is a crucial factor for survival. For example, humans are able to navigate under suddenly changing illumination conditions (driving a car into and out of a tunnel). This is possible by courtesy of adaptation abilities of our sensory organs and pathways. Certainly, adaptation is not confined to a single sense like vision but also affects other senses like audition. Especially the perception of sound source location. Compared to vision, the localization of a sound source in the horizontal plane is a rather complicated task since the location cannot be read out from the receptor surface but needs to be computed. This requires the underlying neural system to calculate differences of the intensity between the two ears which provide a distinct cue for the location of a sound source. Here, adaptation to this cue allows to focus on a specific part of auditory space and thereby facilitates improved localisation abilities. Based on recent findings that suggest that the intensity difference computation is a flexible process with distinct adaptation mechanisms, we developed a neural model that computes the intensity difference to two incoming sound signals. The model comprises a novel mechanism for adaptation to sound source locations and provides a means to investigate underlying neural principles of adaptation and compare their effectivenesses. We demonstrate that due this mechanism the perceptual range is extended and a finer resolution of auditory space is obtained. Results explain the neural basis for adaptation and indicate that the interplay between different adaptation mechanisms facilitate highly precise sound source localization in a wide range of locations.

Predicting neuronal dynamics with a delayed gain control model

Visual neurons respond to static images with specific dynamics: neuronal responses sum sub-additively over time, reduce in amplitude with repeated or sustained stimuli (neuronal adaptation), and are slower at low stimulus contrast. Here, we propose a simple model that predicts these seemingly disparate response patterns observed in a diverse set of measurements-intracranial electrodes in patients, fMRI, and macaque single unit spiking. The model takes a time-varying contrast time course of a stimulus as input, and produces predicted neuronal dynamics as output. Model computation consists of linear filtering, expansive exponentiation, and a divisive gain control. The gain control signal relates to but is slower than the linear signal, and this delay is critical in giving rise to predictions matched to the observed dynamics. Our model is simpler than previously proposed related models, and fitting the model to intracranial EEG data uncovers two regularities across human visual field maps: estimated linear filters (temporal receptive fields) systematically differ across and within visual field maps, and later areas exhibit more rapid and substantial gain control. The model is further generalizable to account for dynamics of contrast-dependent spike rates in macaque V1, and amplitudes of fMRI BOLD in human V1.

The Role of Bottom-Up and Top-Down Cortical Interactions in Adaptation to Natural Scene Statistics

Adaptation is a mechanism by which cortical neurons adjust their responses according to recently viewed stimuli. Visual information is processed in a circuit formed by feedforward (FF) and feedback (FB) synaptic connections of neurons in different cortical layers. Here, the functional role of FF-FB streams and their synaptic dynamics in adaptation to natural stimuli is assessed in psychophysics and neural model. We propose a cortical model which predicts psychophysically observed motion adaptation aftereffects (MAE) after exposure to geometrically distorted natural image sequences. The model comprises direction selective neurons in V1 and MT connected by recurrent FF and FB dynamic synapses. Psychophysically plausible model MAEs were obtained from synaptic changes within neurons tuned to salient direction signals of the broadband natural input. It is conceived that, motion disambiguation by FF-FB interactions is critical to encode this salient information. Moreover, only FF-FB dynamic synapses operating at distinct rates predicted psychophysical MAEs at different adaptation time-scales which could not be accounted for by single rate dynamic synapses in either of the streams. Recurrent FF-FB pathways thereby play a role during adaptation in a natural environment, specifically in inducing multilevel cortical plasticity to salient information and in mediating adaptation at different time-scales.

Pleiotropic action of genetic variation in ZNF804A on brain structure: a meta-analysis of magnetic resonance imaging studies

OBJECTIVE

The zinc finger protein 804A (ZNF804A) gene encodes the protein 804A containing the C2H2 zinc finger structure, which plays an important role in embryonic nerve development and repair. Previous studies have shown a significant association between the ZNF804A genetic variation polymorphism rs1344706 and the risk of schizophrenia and brain structure abnormalities. However, the findings are inconsistent.

MATERIALS AND METHODS

Seventeen studies on structural magnetic resonance imaging (sMRI), with 1,031 schizophrenia patients and 3,416 healthy controls, were included in the meta-analysis. These analyses were performed using Anisotropic Effect-Size Signed Differential Mapping (AES-SDM) software and Comprehensive Meta-Analysis (CMA) software.

RESULTS

rs1344706 risk allele carriers of schizophrenia had increased gray matter in the brain regions including frontal lobe, temporal lobe, and other brain regions, but the carriers of healthy individuals had decreased gray matter and white matter integrity in the frontal lobe, central network, and other brain regions. The results of sensitivity analysis are stable, but publication bias exists in a few analyses of indexes.

CONCLUSION

Abnormalities of brain structure have a strong relationship with ZNF804A gene rs1344706 polymorphism, but the association may be different in healthy individuals and those with mental disorders.

The divisive normalization model of V1 neurons: a comprehensive comparison of physiological data and model predictions

The physiological responses of simple and complex cells in the primary visual cortex (V1) have been studied extensively and modeled at different levels. At the functional level, the divisive normalization model (DNM; Heeger DJ. Vis Neurosci 9: 181-197, 1992) has accounted for a wide range of single-cell recordings in terms of a combination of linear filtering, nonlinear rectification, and divisive normalization. We propose standardizing the formulation of the DNM and implementing it in software that takes static grayscale images as inputs and produces firing rate responses as outputs. We also review a comprehensive suite of 30 empirical phenomena and report a series of simulation experiments that qualitatively replicate dozens of key experiments with a standard parameter set consistent with physiological measurements. This systematic approach identifies novel falsifiable predictions of the DNM. We show how the model simultaneously satisfies the conflicting desiderata of flexibility and falsifiability. Our key idea is that, while adjustable parameters are needed to accommodate the diversity across neurons, they must be fixed for a given individual neuron. This requirement introduces falsifiable constraints when this single neuron is probed with multiple stimuli. We also present mathematical analyses and simulation experiments that explicate some of these constraints.

Biologically Inspired Model for Inference of 3D Shape from Texture

A biologically inspired model architecture for inferring 3D shape from texture is proposed. The model is hierarchically organized into modules roughly corresponding to visual cortical areas in the ventral stream. Initial orientation selective filtering decomposes the input into low-level orientation and spatial frequency representations. Grouping of spatially anisotropic orientation responses builds sketch-like representations of surface shape. Gradients in orientation fields and subsequent integration infers local surface geometry and globally consistent 3D depth. From the distributions in orientation responses summed in frequency, an estimate of the tilt and slant of the local surface can be obtained. The model suggests how 3D shape can be inferred from texture patterns and their image appearance in a hierarchically organized processing cascade along the cortical ventral stream. The proposed model integrates oriented texture gradient information that is encoded in distributed maps of orientation-frequency representations. The texture energy gradient information is defined by changes in the grouped summed normalized orientation-frequency response activity extracted from the textured object image. This activity is integrated by directed fields to generate a 3D shape representation of a complex object with depth ordering proportional to the fields output, with higher activity denoting larger distance in relative depth away from the viewer.

Neural Mechanisms of Cortical Motion Computation Based on a Neuromorphic Sensory System

The visual cortex analyzes motion information along hierarchically arranged visual areas that interact through bidirectional interconnections. This work suggests a bio-inspired visual model focusing on the interactions of the cortical areas in which a new mechanism of feedforward and feedback processing are introduced. The model uses a neuromorphic vision sensor (silicon retina) that simulates the spike-generation functionality of the biological retina. Our model takes into account two main model visual areas, namely V1 and MT, with different feature selectivities. The initial motion is estimated in model area V1 using spatiotemporal filters to locally detect the direction of motion. Here, we adapt the filtering scheme originally suggested by Adelson and Bergen to make it consistent with the spike representation of the DVS. The responses of area V1 are weighted and pooled by area MT cells which are selective to different velocities, i.e. direction and speed. Such feature selectivity is here derived from compositions of activities in the spatio-temporal domain and integrating over larger space-time regions (receptive fields). In order to account for the bidirectional coupling of cortical areas we match properties of the feature selectivity in both areas for feedback processing. For such linkage we integrate the responses over different speeds along a particular preferred direction. Normalization of activities is carried out over the spatial as well as the feature domains to balance the activities of individual neurons in model areas V1 and MT. Our model was tested using different stimuli that moved in different directions. The results reveal that the error margin between the estimated motion and synthetic ground truth is decreased in area MT comparing with the initial estimation of area V1. In addition, the modulated V1 cell activations shows an enhancement of the initial motion estimation that is steered by feedback signals from MT cells.

Reinforcement Learning of Linking and Tracing Contours in Recurrent Neural Networks

The processing of a visual stimulus can be subdivided into a number of stages. Upon stimulus presentation there is an early phase of feedforward processing where the visual information is propagated from lower to higher visual areas for the extraction of basic and complex stimulus features. This is followed by a later phase where horizontal connections within areas and feedback connections from higher areas back to lower areas come into play. In this later phase, image elements that are behaviorally relevant are grouped by Gestalt grouping rules and are labeled in the cortex with enhanced neuronal activity (object-based attention in psychology). Recent neurophysiological studies revealed that reward-based learning influences these recurrent grouping processes, but it is not well understood how rewards train recurrent circuits for perceptual organization. This paper examines the mechanisms for reward-based learning of new grouping rules. We derive a learning rule that can explain how rewards influence the information flow through feedforward, horizontal and feedback connections. We illustrate the efficiency with two tasks that have been used to study the neuronal correlates of perceptual organization in early visual cortex. The first task is called contour-integration and demands the integration of collinear contour elements into an elongated curve. We show how reward-based learning causes an enhancement of the representation of the to-be-grouped elements at early levels of a recurrent neural network, just as is observed in the visual cortex of monkeys. The second task is curve-tracing where the aim is to determine the endpoint of an elongated curve composed of connected image elements. If trained with the new learning rule, neural networks learn to propagate enhanced activity over the curve, in accordance with neurophysiological data. We close the paper with a number of model predictions that can be tested in future neurophysiological and computational studies.

Our experience with the visual world allows us to group image elements that belong to the same perceptual object and to segregate them from other objects and the background. If subjects learn to group contour elements, this experience influences neuronal activity in early visual cortical areas, including the primary visual cortex (V1). Learning presumably depends on alterations in the pattern of connections within and between areas of the visual cortex. However, the processes that control changes in connectivity are not well understood. Here we present the first computational model that can train a neural network to integrate collinear contour elements into elongated curves and to trace a curve through the visual field. The new learning algorithm trains fully recurrent neural networks, provided the connectivity causes the networks to reach a stable state. The model reproduces the behavioral performance of monkeys trained in these tasks and explains the patterns of neuronal activity in the visual cortex that emerge during learning, which is remarkable because the only feedback for the model is a reward for successful trials. We discuss a number of the model predictions that can be tested in future neuroscientific work.

Cognitive functions of intracellular mechanisms for contextual amplification

Evidence for the hypothesis that input to the apical tufts of neocortical pyramidal cells plays a central role in cognition by amplifying their responses to feedforward input is reviewed. Apical tufts are electrically remote from the soma, and their inputs come from diverse sources including direct feedback from higher cortical regions, indirect feedback via the thalamus, and long-range lateral connections both within and between cortical regions. This suggests that input to tuft dendrites may amplify the cell's response to basal inputs that they receive via layer 4 and which have synapses closer to the soma. ERP data supporting this inference is noted. Intracellular studies of apical amplification (AA) and of disamplification by inhibitory interneurons targeted only at tufts are reviewed. Cognitive processes that have been related to them by computational, electrophysiological, and psychopathological studies are then outlined. These processes include: figure-ground segregation and Gestalt grouping; contextual disambiguation in perception and sentence comprehension; priming; winner-take-all competition; attention and working memory; setting the level of consciousness; cognitive control; and learning. It is argued that theories in cognitive neuroscience should not assume that all neurons function as integrate-and-fire point processors, but should use the capabilities of cells with distinct sites of integration for driving and modulatory inputs. Potentially 'unifying' theories that depend upon these capabilities are reviewed. It is concluded that evolution of the primitives of AA and disamplification in neocortex may have extended cognitive capabilities beyond those built from the long-established primitives of excitation, inhibition, and disinhibition.

On event-based optical flow detection

Event-based sensing, i.e., the asynchronous detection of luminance changes, promises low-energy, high dynamic range, and sparse sensing. This stands in contrast to whole image frame-wise acquisition by standard cameras. Here, we systematically investigate the implications of event-based sensing in the context of visual motion, or flow, estimation. Starting from a common theoretical foundation, we discuss different principal approaches for optical flow detection ranging from gradient-based methods over plane-fitting to filter based methods and identify strengths and weaknesses of each class. Gradient-based methods for local motion integration are shown to suffer from the sparse encoding in address-event representations (AER). Approaches exploiting the local plane like structure of the event cloud, on the other hand, are shown to be well suited. Within this class, filter based approaches are shown to define a proper detection scheme which can also deal with the problem of representing multiple motions at a single location (motion transparency). A novel biologically inspired efficient motion detector is proposed, analyzed and experimentally validated. Furthermore, a stage of surround normalization is incorporated. Together with the filtering this defines a canonical circuit for motion feature detection. The theoretical analysis shows that such an integrated circuit reduces motion ambiguity in addition to decorrelating the representation of motion related activations.

Adaptive learning in a compartmental model of visual cortex-how feedback enables stable category learning and refinement

The categorization of real world objects is often reflected in the similarity of their visual appearances. Such categories of objects do not necessarily form disjunct sets of objects, neither semantically nor visually. The relationship between categories can often be described in terms of a hierarchical structure. For instance, tigers and leopards build two separate mammalian categories, both of which are subcategories of the category Felidae. In the last decades, the unsupervised learning of categories of visual input stimuli has been addressed by numerous approaches in machine learning as well as in computational neuroscience. However, the question of what kind of mechanisms might be involved in the process of subcategory learning, or category refinement, remains a topic of active investigation. We propose a recurrent computational network architecture for the unsupervised learning of categorial and subcategorial visual input representations. During learning, the connection strengths of bottom-up weights from input to higher-level category representations are adapted according to the input activity distribution. In a similar manner, top-down weights learn to encode the characteristics of a specific stimulus category. Feedforward and feedback learning in combination realize an associative memory mechanism, enabling the selective top-down propagation of a category's feedback weight distribution. We suggest that the difference between the expected input encoded in the projective field of a category node and the current input pattern controls the amplification of feedforward-driven representations. Large enough differences trigger the recruitment of new representational resources and the establishment of additional (sub-) category representations. We demonstrate the temporal evolution of such learning and show how the proposed combination of an associative memory with a modulatory feedback integration successfully establishes category and subcategory representations.