Recent neurophysiological studies have revealed that the early visual cortex can rapidly learn global image context, as evidenced by a sparsification of population responses and a reduction in mean activity when exposed to familiar versus novel image contexts. This phenomenon has been attributed primarily to local recurrent interactions, rather than changes in feedforward or feedback pathways-supported by both empirical findings and circuit-level modeling. Recurrent neural circuits capable of simulating these effects have been shown to reshape the geometry of neural manifolds, enhancing robustness and invariance to irrelevant variations. In this study, we employ a Vision Transformer (ViT)-based autoencoder to investigate, from a functional perspective, how familiarity training can induce sensitivity to global context in the early layers of a deep neural network. We hypothesize that rapid learning operates via fast weights, which encode transient or short-term memory traces, and we explore the use of Low-Rank Adaptation (LoRA) to implement such fast weights within each Transformer layer. Our results show that: (1) The proposed ViT-based autoencoder's self-attention circuit is performing a manifold transform similar to a neural circuit developed for modeling the familiarity effect. (2) Familiarity training induces alignment of latent representation in early layers with the top layer that contains global context information. (3) Familiarity training makes self-attention pay attention to a broader scope details in the remembered image context, rather than just the critical features for object recognition. (4) These effects are significantly amplified by the incorporation of LoRA-based fast weights. Together, these findings suggest that familiarity training can introduce global sensitivity to earlier layers in a hierarchical network, and that a hybrid fast-and-slow weight architecture may provide a viable computational model for studying the functional consequences of rapid global context learning in the brain.

David Marr's seminal theory of human perception stipulates that visual processing is a multi-stage process, prioritizing the derivation of boundary and surface properties before forming semantic object representations. In contrast, contrastive representation learning frameworks typically bypass this explicit multi-stage approach, defining their objective as the direct learning of a semantic representation space for objects. While effective in general contexts, this approach sacrifices the inductive biases of vision, leading to slower convergence speed and learning shortcut resulting in texture bias. In this work, we demonstrate that leveraging Marr's multi-stage theory-by first constructing boundary and surface-level representations using perceptual constructs from early visual processing stages and subsequently training for object semantics-leads to 2x faster convergence on ResNet18, improved final representations on semantic segmentation, depth estimation, and object recognition, and enhanced robustness and out-of-distribution capability. Together, we propose a pretraining stage before the general contrastive representation pretraining to further enhance the final representation quality and reduce the overall convergence time via inductive bias from human vision systems.

Convolutional neural networks (CNNs) have been shown to be state-of-the-art models for visual cortical neurons. Cortical neurons in the primary visual cortex are sensitive to contextual information mediated by extensive horizontal and feedback connections. Standard CNNs integrate global contextual information to model contextual modulation via two mechanisms: successive convolutions and a fully connected readout layer. In this paper, we find that self-attention (SA), an implementation of non-local network mechanisms, can improve neural response predictions over parameter-matched CNNs in two key metrics: tuning curve correlation and peak tuning. We introduce peak tuning as a metric to evaluate a model's ability to capture a neuron's top feature preference. We factorize networks to assess each context mechanism, revealing that information in the local receptive field is most important for modeling overall tuning, but surround information is critically necessary for characterizing the tuning peak. We find that self-attention can replace posterior spatial-integration convolutions when learned incrementally, and is further enhanced in the presence of a fully connected readout layer, suggesting that the two context mechanisms are complementary. Finally, we find that decomposing receptive field learning and contextual modulation learning in an incremental manner may be an effective and robust mechanism for learning surround-center interactions.

A ubiquitous phenomenon observed along the ventral stream of the primate hierarchical visual system is the suppression of neural responses to familiar stimuli at the population level. The observation of the suppression of the neural response in the early visual cortex (V1 and V2) to familiar stimuli that are multiple times larger in size than the receptive fields of individual neurons implicates the plausible development of recurrent circuits for encoding these global stimuli. In this work, we investigated the neural mechanisms of familiarity suppression and showed that a recurrent neural circuit based on Hebbian learning, consisting of neurons with small and local receptive fields, can develop to encode specific global familiar stimuli robustly as a result of familiarity training. We proposed that the learned recurrent circuit implements a manifold transform. The recurrent circuit compresses the dimensions of nuisance variations of a familiar image in the neural response manifold relative to the dimensions for discriminating different familiar stimuli, resulting in increased robustness of the global stimulus representation against noise and other irrelevant perturbations. We demonstrate that a recurrent circuit implements the manifold transform using a mixed strategy of locally linear and globally nonlinear computations, where the local linear computation selectively redistributes recurrent gain to enhance concept discrimination. These results provide testable predictions for neurophysiological experiments.

Current deep-learning models for object recognition are known to be heavily biased toward texture. In contrast, human visual systems are known to be biased toward shape and structure. What could be the design principles in human visual systems that led to this difference? How could we introduce more shape bias into the deep learning models? In this paper, we report that sparse coding, a ubiquitous principle in the brain, can in itself introduce shape bias into the network. We found that enforcing the sparse coding constraint using a non-differential Top-K operation can lead to the emergence of structural encoding in neurons in convolutional neural networks, resulting in a smooth decomposition of objects into parts and subparts and endowing the networks with shape bias. We demonstrated this emergence of shape bias and its functional benefits for different network structures with various datasets. For object recognition convolutional neural networks, the shape bias leads to greater robustness against style and pattern change distraction. For the image synthesis generative adversary networks, the emerged shape bias leads to more coherent and decomposable structures in the synthesized images. Ablation studies suggest that sparse codes tend to encode structures, whereas the more distributed codes tend to favor texture. Our code is host at the github repository: https://topk-shape-bias.github.io/

Recent advances in deep generative models have led to immense progress in 3D shape synthesis. While existing models are able to synthesize shapes represented as voxels, point-clouds, or implicit functions, these methods only indirectly enforce the plausibility of the final 3D shape surface. Here we present a 3D shape synthesis framework (SurfGen) that directly applies adversarial training to the object surface. Our approach uses a differentiable spherical projection layer to capture and represent the explicit zero isosurface of an implicit 3D generator as functions defined on the unit sphere. By processing the spherical representation of 3D object surfaces with a spherical CNN in an adversarial setting, our generator can better learn the statistics of natural shape surfaces. We evaluate our model on large-scale shape datasets, and demonstrate that the end-to-end trained model is capable of generating high fidelity 3D shapes with diverse topology.

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.

This study provides a normative theory for how Bayesian causal inference can be implemented in neural circuits. In both cognitive processes such as causal reasoning and perceptual inference such as cue integration, the nervous systems need to choose different models representing the underlying causal structures when making inferences on external stimuli. In multisensory processing, for example, the nervous system has to choose whether to integrate or segregate inputs from different sensory modalities to infer the sensory stimuli, based on whether the inputs are from the same or different sources. Making this choice is a model selection problem requiring the computation of Bayes factor, the ratio of likelihoods between the integration and the segregation models. In this paper, we consider the causal inference in multisensory processing and propose a novel generative model based on neural population code that takes into account both stimulus feature and stimulus reliability in the inference. In the case of circular variables such as heading direction, our normative theory yields an analytical solution for computing the Bayes factor, with a clear geometric interpretation, which can be implemented by simple additive mechanisms with neural population code. Numerical simulation shows that the tunings of the neurons computing Bayes factor are consistent with the "opposite neurons" discovered in dorsal medial superior temporal (MSTd) and the ventral intraparietal (VIP) areas for visual-vestibular processing. This study illuminates a potential neural mechanism for causal inference in the brain.

We propose using the statistical measurement of the sample skewness of the distribution of mean firing rates of a tuning curve to quantify sharpness of tuning. For some features, like binocular disparity, tuning curves are best described by relatively complex and sometimes diverse functions, making it difficult to quantify sharpness with a single function and parameter. Skewness provides a robust nonparametric measure of tuning curve sharpness that is invariant with respect to the mean and variance of the tuning curve and is straightforward to apply to a wide range of tuning, including simple orientation tuning curves and complex object tuning curves that often cannot even be described parametrically. Because skewness does not depend on a specific model or function of tuning, it is especially appealing to cases of sharpening where recurrent interactions among neurons produce sharper tuning curves that deviate in a complex manner from the feedforward function of tuning. Since tuning curves for all neurons are not typically well described by a single parametric function, this model independence additionally allows skewness to be applied to all recorded neurons, maximizing the statistical power of a set of data. We also compare skewness with other nonparametric measures of tuning curve sharpness and selectivity. Compared to these other nonparametric measures tested, skewness is best used for capturing the sharpness of multimodal tuning curves defined by narrow peaks (maximum) and broad valleys (minima). Finally, we provide a more formal definition of sharpness using a shape-based information gain measure and derive and show that skewness is correlated with this definition.

Integrating generative models and discriminative models in a hybrid scheme has shown some success in recognition tasks. In such scheme, generative models are used to derive feature maps for outputting a set of fixed length features that are used by discriminative models to perform classification. In this paper, we present a method, called posterior divergence, to derive feature maps from the log likelihood function implied in the incremental expectation-maximization algorithm. These feature maps evaluate a sample in three complementary measures: (1) how much the sample affects the model; (2) how well the sample fits the model; (3) how uncertain the fit is. We prove that the linear classification error rate using the outputs of the derived feature maps is at least as low as that of plug-in estimation. We present efficient algorithms for computing these feature maps for semi-supervised learning and supervised learning. We evaluate the proposed method on three typical applications, i.e. scene recognition, face and non-face classification and protein sequence analysis, and demonstrate improvements over related methods.

Activity of a neuron, even in the early sensory areas, is not simply a function of its local receptive field or tuning properties, but depends on global context of the stimulus, as well as the neural context. This suggests the activity of the surrounding neurons and global brain states can exert considerable influence on the activity of a neuron. In this paper we implemented an L1 regularized point process model to assess the contribution of multiple factors to the firing rate of many individual units recorded simultaneously from V1 with a 96-electrode Utah" array. We found that the spikes of surrounding neurons indeed provide strong predictions of a neuron's response, in addition to the neuron's receptive field transfer function. We also found that the same spikes could be accounted for with the local field potentials, a surrogate measure of global network states. This work shows that accounting for network fluctuations can improve estimates of single trial firing rate and stimulus-response transfer functions."

Although there has been substantial progress in understanding the neuro- physiological mechanisms of stereopsis, how neurons interact in a network during stereo computation remains unclear. Computational models on stereopsis suggest local competition and long-range cooperation are impor- tant for resolving ambiguity during stereo matching. To test these predic- tions, we simultaneously recorded from multiple neurons in V1 of awake, behaving macaques while presenting surfaces of different depths rendered in dynamic random dot stereograms. We found that the interaction between pairs of neurons was a function of similarity in receptive fields, as well as of the input stimulus. Neurons coding the same depth experienced common inhibition early in their responses for stimuli presented at their non- preferred disparities. They experienced mutual facilitation later in their re- sponses for stimulation at their preferred disparity. These findings are con- sistent with a local competition mechanism that first removes gross mis- matches, and a global cooperative mechanism that further refines depth es- timates.

This paper explores the statistical relationship between natural images and their underlying range (depth) images. We look at how this relationship changes over scale, and how this information can be used to enhance low resolution range data using a full resolution intensity image. Based on our findings, we propose an extension to an existing technique known as shape recipes [3], and the success of the two methods are compared using images and laser scans of real scenes. Our extension is shown to provide a two-fold improvement over the current method. Furthermore, we demonstrate that ideal linear shape-from-shading filters, when learned from natural scenes, may derive even more strength from shadow cues than from the traditional linear-Lambertian shading cues.

Decoding is a strategy that allows us to assess the amount of information neurons can provide about certain aspects of the visual scene. In this study, we develop a method based on Bayesian sequential updating and the particle filtering algorithm to decode the activity of V1 neurons in awake monkeys. A distinction in our method is the use of Volterra ker- nels to filter the particles, which live in a high dimensional space. This parametric Bayesian decoding scheme is compared to the optimal linear decoder and is shown to work consistently better than the linear optimal decoder. Interestingly, our results suggest that for decoding in real time, spike trains of as few as 10 independent but similar neurons would be sufficient for decoding a critical scene variable in a particular class of visual stimuli. The reconstructed variable can predict the neural activity about as well as the actual signal with respect to the Volterra kernels.

This paper extends to two dimensions the frame criterion developed by Daubechies for one-dimensional wavelets, and it computes the frame bounds for the particular case of 2D Gabor wavelets. Completeness criteria for 2D Gabor image representations are important because of their increasing role in many computer vision applications and also in modeling biological vision, since recent neurophysiological evidence from the visual cortex of mammalian brains suggests that the filter response profiles of the main class of linearly-responding cortical neurons (called simple cells) are best modeled as a family of self-similar 2D Gabor wavelets. We therefore derive the conditions under which a set of continuous 2D Gabor wavelets will provide a complete representation of any image, and we also find self-similar wavelet parametrization which allow stable reconstruction by summation as though the wavelets formed an orthonormal basis. Approximating a "tight frame" generates redundancy which allows low-resolution neural responses to represent high-resolution images.

We present a novel statistical and variational approach to image segmentation based on a new algorithm named region competition. This algorithm is derived by minimizing a generalized Bayes/MDL (Minimum Description Length) criterion using the variational principle. We show that existing techniques in early vision such as, snake/balloon models, region growing, and Bayes/MDL are addressing different aspects of the same problem and they can be unified within a common statistical framework which combines their advantages. We analyze how to optimize the precision of the resulting boundary location by studying the statistical properties of the region competition algorithm and discuss what are good initial conditions for the algorithm. Our method is generalized to color and texture segmentation and is demonstrated on grey level images, color images and texture images.>

We generalized Daubechies' (see IEEE Transactions on Information Theory, vol.36, no.5, p.961-1004, 1990) results on 1D wavelets to 2D, and elucidated the conditions in which a family of Gabor wavelets will provide a complete representation of an image. We found that the phase space (scale, orientation, x, y) sampling density of the striate cortical simple cells is sufficiently dense to form a tight wavelet frame. This tight frame approximates the continuous wavelet transform, and its redundancy allows high resolution information to be represented in the low-resolution firings of the simple cells. We presented some experimental results to demonstrate this idea.>