Is slowness a learning principle of the visual cortex?

Slow feature analysis is an algorithm for extracting slowly varying features from a quickly varying signal. It has been shown in network simulations on one-dimensional stimuli that visual invariances to shift and other transformations can be learned in an unsupervised fashion based on slow feature analysis. More recently, we have shown that slow feature analysis applied to image sequences generated from natural images using a range of spatial transformations results in units that share many properties with complex and hypercomplex cells of the primary visual cortex. We find cells responsive to Gabor stimuli with phase invariance, sharpened or widened orientation or frequency tuning, secondary response lobes, end-stopping, and cells selective for direction of motion. These results indicate that slowness may be an important principle of self-organization in the visual cortex.     

Learning receptive fields using predictive feedback.

Previously, it was suggested that feedback connections from higher- to lower-level areas carry predictions of lower-level neural activities, whereas feedforward connections carry the residual error between the predictions and the actual lower-level activities [Rao, R.P.N., Ballard, D.H., 1999. Nature Neuroscience 2, 79-87.]. A computational model implementing the hypothesis learned simple cell receptive fields when exposed to natural images. Here, we use predictive feedback to explain tuning properties in medial superior temporal area (MST). We implement the hypothesis using a new, biologically plausible, algorithm based on matching pursuit, which retains all the features of the previous implementation, including its ability to efficiently encode input. When presented with natural images, the model developed receptive field properties as found in primary visual cortex. In addition, when exposed to visual motion input resulting from movements through space, the model learned receptive field properties resembling those in MST. These results corroborate the idea that predictive feedback is a general principle used by the visual system to efficiently encode natural input.     

Sparse Coding Models Can Exhibit Decreasing Sparseness while Learning Sparse Codes for Natural Images

The sparse coding hypothesis has enjoyed much success in predicting response properties of simple cells in primary visual cortex (V1) based solely on the statistics of natural scenes. In typical sparse coding models, model neuron activities and receptive fields are optimized to accurately represent input stimuli using the least amount of neural activity. As these networks develop to represent a given class of stimulus, the receptive fields are refined so that they capture the most important stimulus features. Intuitively, this is expected to result in sparser network activity over time. Recent experiments, however, show that stimulus-evoked activity in ferret V1 becomes less sparse during development, presenting an apparent challenge to the sparse coding hypothesis. Here we demonstrate that some sparse coding models, such as those employing homeostatic mechanisms on neural firing rates, can exhibit decreasing sparseness during learning, while still achieving good agreement with mature V1 receptive field shapes and a reasonably sparse mature network state. We conclude that observed developmental trends do not rule out sparseness as a principle of neural coding per se: a mature network can perform sparse coding even if sparseness decreases somewhat during development. To make comparisons between model and physiological receptive fields, we introduce a new nonparametric method for comparing receptive field shapes using image registration techniques.     

Invariance of visual operations at the level of receptive fields

The brain is able to maintain a stable perception although the visual stimuli vary substantially on the retina due to geometric transformations and lighting variations in the environment. This paper presents a theory for achieving basic invariance properties already at the level of receptive fields. Specifically, the presented framework comprises (i) local scaling transformations caused by objects of different size and at different distances to the observer, (ii) locally linearized image deformations caused by variations in the viewing direction in relation to the object, (iii) locally linearized relative motions between the object and the observer and (iv) local multiplicative intensity transformations caused by illumination variations. The receptive field model can be derived by necessity from symmetry properties of the environment and leads to predictions about receptive field profiles in good agreement with receptive field profiles measured by cell recordings in mammalian vision. Indeed, the receptive field profiles in the retina, LGN and V1 are close to ideal to what is motivated by the idealized requirements. By complementing receptive field measurements with selection mechanisms over the parameters in the receptive field families, it is shown how true invariance of receptive field responses can be obtained under scaling transformations, affine transformations and Galilean transformations. Thereby, the framework provides a mathematically well-founded and biologically plausible model for how basic invariance properties can be achieved already at the level of receptive fields and support invariant recognition of objects and events under variations in viewpoint, retinal size, object motion and illumination. The theory can explain the different shapes of receptive field profiles found in biological vision, which are tuned to different sizes and orientations in the image domain as well as to different image velocities in space-time, from a requirement that the visual system should be invariant to the natural types of image transformations that occur in its environment.     

Discriminative Learning of Receptive Fields from Responses to Non-Gaussian Stimulus Ensembles

Analysis of sensory neurons'' processing characteristics requires simultaneous measurement of presented stimuli and concurrent spike responses. The functional transformation from high-dimensional stimulus space to the binary space of spike and non-spike responses is commonly described with linear-nonlinear models, whose linear filter component describes the neuron''s receptive field. From a machine learning perspective, this corresponds to the binary classification problem of discriminating spike-eliciting from non-spike-eliciting stimulus examples. The classification-based receptive field (CbRF) estimation method proposed here adapts a linear large-margin classifier to optimally predict experimental stimulus-response data and subsequently interprets learned classifier weights as the neuron''s receptive field filter. Computational learning theory provides a theoretical framework for learning from data and guarantees optimality in the sense that the risk of erroneously assigning a spike-eliciting stimulus example to the non-spike class (and vice versa) is minimized. Efficacy of the CbRF method is validated with simulations and for auditory spectro-temporal receptive field (STRF) estimation from experimental recordings in the auditory midbrain of Mongolian gerbils. Acoustic stimulation is performed with frequency-modulated tone complexes that mimic properties of natural stimuli, specifically non-Gaussian amplitude distribution and higher-order correlations. Results demonstrate that the proposed approach successfully identifies correct underlying STRFs, even in cases where second-order methods based on the spike-triggered average (STA) do not. Applied to small data samples, the method is shown to converge on smaller amounts of experimental recordings and with lower estimation variance than the generalized linear model and recent information theoretic methods. Thus, CbRF estimation may prove useful for investigation of neuronal processes in response to natural stimuli and in settings where rapid adaptation is induced by experimental design.     

Slowness: an objective for spike-timing-dependent plasticity?

Our nervous system can efficiently recognize objects in spite of changes in contextual variables such as perspective or lighting conditions. Several lines of research have proposed that this ability for invariant recognition is learned by exploiting the fact that object identities typically vary more slowly in time than contextual variables or noise. Here, we study the question of how this "temporal stability" or "slowness" approach can be implemented within the limits of biologically realistic spike-based learning rules. We first show that slow feature analysis, an algorithm that is based on slowness, can be implemented in linear continuous model neurons by means of a modified Hebbian learning rule. This approach provides a link to the trace rule, which is another implementation of slowness learning. Then, we show analytically that for linear Poisson neurons, slowness learning can be implemented by spike-timing-dependent plasticity (STDP) with a specific learning window. By studying the learning dynamics of STDP, we show that for functional interpretations of STDP, it is not the learning window alone that is relevant but rather the convolution of the learning window with the postsynaptic potential. We then derive STDP learning windows that implement slow feature analysis and the "trace rule." The resulting learning windows are compatible with physiological data both in shape and timescale. Moreover, our analysis shows that the learning window can be split into two functionally different components that are sensitive to reversible and irreversible aspects of the input statistics, respectively. The theory indicates that irreversible input statistics are not in favor of stable weight distributions but may generate oscillatory weight dynamics. Our analysis offers a novel interpretation for the functional role of STDP in physiological neurons.     

A sparse coding model with synaptically local plasticity and spiking neurons can account for the diverse shapes of V1 simple cell receptive fields

Sparse coding algorithms trained on natural images can accurately predict the features that excite visual cortical neurons, but it is not known whether such codes can be learned using biologically realistic plasticity rules. We have developed a biophysically motivated spiking network, relying solely on synaptically local information, that can predict the full diversity of V1 simple cell receptive field shapes when trained on natural images. This represents the first demonstration that sparse coding principles, operating within the constraints imposed by cortical architecture, can successfully reproduce these receptive fields. We further prove, mathematically, that sparseness and decorrelation are the key ingredients that allow for synaptically local plasticity rules to optimize a cooperative, linear generative image model formed by the neural representation. Finally, we discuss several interesting emergent properties of our network, with the intent of bridging the gap between theoretical and experimental studies of visual cortex.     

A computational theory of visual receptive fields

A receptive field constitutes a region in the visual field where a visual cell or a visual operator responds to visual stimuli. This paper presents a theory for what types of receptive field profiles can be regarded as natural for an idealized vision system, given a set of structural requirements on the first stages of visual processing that reflect symmetry properties of the surrounding world. These symmetry properties include (i) covariance properties under scale changes, affine image deformations, and Galilean transformations of space–time as occur for real-world image data as well as specific requirements of (ii) temporal causality implying that the future cannot be accessed and (iii) a time-recursive updating mechanism of a limited temporal buffer of the past as is necessary for a genuine real-time system. Fundamental structural requirements are also imposed to ensure (iv) mutual consistency and a proper handling of internal representations at different spatial and temporal scales. It is shown how a set of families of idealized receptive field profiles can be derived by necessity regarding spatial, spatio-chromatic, and spatio-temporal receptive fields in terms of Gaussian kernels, Gaussian derivatives, or closely related operators. Such image filters have been successfully used as a basis for expressing a large number of visual operations in computer vision, regarding feature detection, feature classification, motion estimation, object recognition, spatio-temporal recognition, and shape estimation. Hence, the associated so-called scale-space theory constitutes a both theoretically well-founded and general framework for expressing visual operations. There are very close similarities between receptive field profiles predicted from this scale-space theory and receptive field profiles found by cell recordings in biological vision. Among the family of receptive field profiles derived by necessity from the assumptions, idealized models with very good qualitative agreement are obtained for (i) spatial on-center/off-surround and off-center/on-surround receptive fields in the fovea and the LGN, (ii) simple cells with spatial directional preference in V1, (iii) spatio-chromatic double-opponent neurons in V1, (iv) space–time separable spatio-temporal receptive fields in the LGN and V1, and (v) non-separable space–time tilted receptive fields in V1, all within the same unified theory. In addition, the paper presents a more general framework for relating and interpreting these receptive fields conceptually and possibly predicting new receptive field profiles as well as for pre-wiring covariance under scaling, affine, and Galilean transformations into the representations of visual stimuli. This paper describes the basic structure of the necessity results concerning receptive field profiles regarding the mathematical foundation of the theory and outlines how the proposed theory could be used in further studies and modelling of biological vision. It is also shown how receptive field responses can be interpreted physically, as the superposition of relative variations of surface structure and illumination variations, given a logarithmic brightness scale, and how receptive field measurements will be invariant under multiplicative illumination variations and exposure control mechanisms.     

Slowness: An Objective for Spike-Timing–Dependent Plasticity?

Our nervous system can efficiently recognize objects in spite of changes in contextual variables such as perspective or lighting conditions. Several lines of research have proposed that this ability for invariant recognition is learned by exploiting the fact that object identities typically vary more slowly in time than contextual variables or noise. Here, we study the question of how this “temporal stability” or “slowness” approach can be implemented within the limits of biologically realistic spike-based learning rules. We first show that slow feature analysis, an algorithm that is based on slowness, can be implemented in linear continuous model neurons by means of a modified Hebbian learning rule. This approach provides a link to the trace rule, which is another implementation of slowness learning. Then, we show analytically that for linear Poisson neurons, slowness learning can be implemented by spike-timing–dependent plasticity (STDP) with a specific learning window. By studying the learning dynamics of STDP, we show that for functional interpretations of STDP, it is not the learning window alone that is relevant but rather the convolution of the learning window with the postsynaptic potential. We then derive STDP learning windows that implement slow feature analysis and the “trace rule.” The resulting learning windows are compatible with physiological data both in shape and timescale. Moreover, our analysis shows that the learning window can be split into two functionally different components that are sensitive to reversible and irreversible aspects of the input statistics, respectively. The theory indicates that irreversible input statistics are not in favor of stable weight distributions but may generate oscillatory weight dynamics. Our analysis offers a novel interpretation for the functional role of STDP in physiological neurons.     

Receptive Field Vectors of Genetically-Identified Retinal Ganglion Cells Reveal Cell-Type-Dependent Visual Functions

Sensory stimuli are encoded by diverse kinds of neurons but the identities of the recorded neurons that are studied are often unknown. We explored in detail the firing patterns of eight previously defined genetically-identified retinal ganglion cell (RGC) types from a single transgenic mouse line. We first introduce a new technique of deriving receptive field vectors (RFVs) which utilises a modified form of mutual information (“Quadratic Mutual Information”). We analysed the firing patterns of RGCs during presentation of short duration (~10 second) complex visual scenes (natural movies). We probed the high dimensional space formed by the visual input for a much smaller dimensional subspace of RFVs that give the most information about the response of each cell. The new technique is very efficient and fast and the derivation of novel types of RFVs formed by the natural scene visual input was possible even with limited numbers of spikes per cell. This approach enabled us to estimate the ''visual memory'' of each cell type and the corresponding receptive field area by calculating Mutual Information as a function of the number of frames and radius. Finally, we made predictions of biologically relevant functions based on the RFVs of each cell type. RGC class analysis was complemented with results for the cells’ response to simple visual input in the form of black and white spot stimulation, and their classification on several key physiological metrics. Thus RFVs lead to predictions of biological roles based on limited data and facilitate analysis of sensory-evoked spiking data from defined cell types.     

Sparsity and Compressed Coding in Sensory Systems

Considering that many natural stimuli are sparse, can a sensory system evolve to take advantage of this sparsity? We explore this question and show that significant downstream reductions in the numbers of neurons transmitting stimuli observed in early sensory pathways might be a consequence of this sparsity. First, we model an early sensory pathway using an idealized neuronal network comprised of receptors and downstream sensory neurons. Then, by revealing a linear structure intrinsic to neuronal network dynamics, our work points to a potential mechanism for transmitting sparse stimuli, related to compressed-sensing (CS) type data acquisition. Through simulation, we examine the characteristics of networks that are optimal in sparsity encoding, and the impact of localized receptive fields beyond conventional CS theory. The results of this work suggest a new network framework of signal sparsity, freeing the notion from any dependence on specific component-space representations. We expect our CS network mechanism to provide guidance for studying sparse stimulus transmission along realistic sensory pathways as well as engineering network designs that utilize sparsity encoding.     

Adaptation to stimulus contrast and correlations during natural visual stimulation

In this study, we characterize the adaptation of neurons in the cat lateral geniculate nucleus to changes in stimulus contrast and correlations. By comparing responses to high- and low-contrast natural scene movie and white noise stimuli, we show that an increase in contrast or correlations results in receptive fields with faster temporal dynamics and stronger antagonistic surrounds, as well as decreases in gain and selectivity. We also observe contrast- and correlation-induced changes in the reliability and sparseness of neural responses. We find that reliability is determined primarily by processing in the receptive field (the effective contrast of the stimulus), while sparseness is determined by the interactions between several functional properties. These results reveal a number of adaptive phenomena and suggest that adaptation to stimulus contrast and correlations may play an important role in visual coding in a dynamic natural environment.     

Specific Relationship between Excitatory Inputs and Climbing Fiber Receptive Fields in Deep Cerebellar Nuclear Neurons

Many mossy fiber pathways to the neurons of the deep cerebellar nucleus (DCN) originate from the spinal motor circuitry. For cutaneously activated spinal neurons, the receptive field is a tag indicating the specific motor function the spinal neuron has. Similarly, the climbing fiber receptive field of the DCN neuron reflects the specific motor output function of the DCN neuron. To explore the relationship between the motor information the DCN neuron receives and the output it issues, we made patch clamp recordings of DCN cell responses to tactile skin stimulation in the forelimb region of the anterior interposed nucleus in vivo. The excitatory responses were organized according to a general principle, in which the DCN cell responses became stronger the closer the skin site was located to its climbing fiber receptive field. The findings represent a novel functional principle of cerebellar connectivity, with crucial importance for our understanding of the function of the cerebellum in movement coordination.     

Complexity and diversity in sparse code priors improve receptive field characterization of Macaque V1 neurons

System identification techniques—projection pursuit regression models (PPRs) and convolutional neural networks (CNNs)—provide state-of-the-art performance in predicting visual cortical neurons’ responses to arbitrary input stimuli. However, the constituent kernels recovered by these methods are often noisy and lack coherent structure, making it difficult to understand the underlying component features of a neuron’s receptive field. In this paper, we show that using a dictionary of diverse kernels with complex shapes learned from natural scenes based on efficient coding theory, as the front-end for PPRs and CNNs can improve their performance in neuronal response prediction as well as algorithmic data efficiency and convergence speed. Extensive experimental results also indicate that these sparse-code kernels provide important information on the component features of a neuron’s receptive field. In addition, we find that models with the complex-shaped sparse code front-end are significantly better than models with a standard orientation-selective Gabor filter front-end for modeling V1 neurons that have been found to exhibit complex pattern selectivity. We show that the relative performance difference due to these two front-ends can be used to produce a sensitive metric for detecting complex selectivity in V1 neurons.     

Are V1 Simple Cells Optimized for Visual Occlusions? A Comparative Study

Simple cells in primary visual cortex were famously found to respond to low-level image components such as edges. Sparse coding and independent component analysis (ICA) emerged as the standard computational models for simple cell coding because they linked their receptive fields to the statistics of visual stimuli. However, a salient feature of image statistics, occlusions of image components, is not considered by these models. Here we ask if occlusions have an effect on the predicted shapes of simple cell receptive fields. We use a comparative approach to answer this question and investigate two models for simple cells: a standard linear model and an occlusive model. For both models we simultaneously estimate optimal receptive fields, sparsity and stimulus noise. The two models are identical except for their component superposition assumption. We find the image encoding and receptive fields predicted by the models to differ significantly. While both models predict many Gabor-like fields, the occlusive model predicts a much sparser encoding and high percentages of ‘globular’ receptive fields. This relatively new center-surround type of simple cell response is observed since reverse correlation is used in experimental studies. While high percentages of ‘globular’ fields can be obtained using specific choices of sparsity and overcompleteness in linear sparse coding, no or only low proportions are reported in the vast majority of studies on linear models (including all ICA models). Likewise, for the here investigated linear model and optimal sparsity, only low proportions of ‘globular’ fields are observed. In comparison, the occlusive model robustly infers high proportions and can match the experimentally observed high proportions of ‘globular’ fields well. Our computational study, therefore, suggests that ‘globular’ fields may be evidence for an optimal encoding of visual occlusions in primary visual cortex.     

Inhomogeneous retino-cortical mapping is supported and stabilized with correlation-learning during self-motion

In primates, the area of primary visual cortex representing a fixed area of visual space decreases with increasing eccentricity. We identify visual situations to which this inhomogeneous retino-cortical mapping is well adapted and study their relevance during natural vision and development. We assume that cortical activations caused by stationary objects during self-motion along the direction of gaze travel on average with constant speed across the cortical surface, independent of retinal eccentricity. This is the case if the distribution of objects corresponds to an ellipsoid with the observer in its center. We apply the resulting flow field to train a simple network of pulse coding neurons with Hebbian learning and demonstrate that the density of learned receptive field centers is in close agreement with primate retino-cortical magnification. In addition, the model reproduces the increase of receptive field size and the decrease of its peak sensitivity with increasing eccentricity. Our results suggest that self-motion may have played an important role in the evolution of the visual system and that cortical magnification can be refined and stabilized by Hebbian learning mechanisms in ontogenesis under natural viewing conditions.     

Transformation of Stimulus Correlations by the Retina

Redundancies and correlations in the responses of sensory neurons may seem to waste neural resources, but they can also carry cues about structured stimuli and may help the brain to correct for response errors. To investigate the effect of stimulus structure on redundancy in retina, we measured simultaneous responses from populations of retinal ganglion cells presented with natural and artificial stimuli that varied greatly in correlation structure; these stimuli and recordings are publicly available online. Responding to spatio-temporally structured stimuli such as natural movies, pairs of ganglion cells were modestly more correlated than in response to white noise checkerboards, but they were much less correlated than predicted by a non-adapting functional model of retinal response. Meanwhile, responding to stimuli with purely spatial correlations, pairs of ganglion cells showed increased correlations consistent with a static, non-adapting receptive field and nonlinearity. We found that in response to spatio-temporally correlated stimuli, ganglion cells had faster temporal kernels and tended to have stronger surrounds. These properties of individual cells, along with gain changes that opposed changes in effective contrast at the ganglion cell input, largely explained the pattern of pairwise correlations across stimuli where receptive field measurements were possible.     

Common reducing subspace model and network alternation analysis

Motivated by brain connectivity analysis and many other network data applications, we study the problem of estimating covariance and precision matrices and their differences across multiple populations. We propose a common reducing subspace model that leads to substantial dimension reduction and efficient parameter estimation. We explicitly quantify the efficiency gain through an asymptotic analysis. Our method is built upon and further extends a nascent technique, the envelope model, which adopts a generalized sparsity principle. This distinguishes our proposal from most xisting covariance and precision estimation methods that assume element‐wise sparsity. Moreover, unlike most existing solutions, our method can naturally handle both covariance and precision matrices in a unified way, and work with matrix‐valued data. We demonstrate the efficacy of our method through intensive simulations, and illustrate the method with an autism spectrum disorder data analysis.     

Engineering a versatile tandem repeat-type α2-6sialic acid-binding lectin

Previously, we developed an α2-6-sialic acid (Sia)-specific lectin (SRC) starting from an R-type galactose-specific lectin C-terminal domain. However, it showed relatively low affinity because of its monovalency. Here, we engineered a tandem repeat construct (SRC2) showing substantial affinity for α2,6-sialylated N-glycans (in the order of 10⁻⁶ M in K_d), almost comparable to a natural α2-6Sia-specific lectin from Sambucus sieboldiana (SSA). Notably, its binding to branched N-glycans was found to be more selective than SSA. Nevertheless, SRC2 showed no apparent hemagglutinating activity, while it exerted strong erythrocyte-binding activity. This unique feature will help flow cytometry analysis, where usual lectins including SSA agglutinate cells. Some other biochemical properties investigated for SRC2, e.g., high productivity in bacteria and easy release of captured glycoproteins with lactose have demonstrated versatility of this mutant protein as a powerful tool for sialoglycomics.