Constrained inference in sparse coding reproduces contextual effects and predicts laminar neural dynamics

Abstract

When probed with complex stimuli that extend beyond their classical receptive field, neurons in primary visual cortex display complex and non-linear response characteristics. Sparse coding models reproduce some of the observed contextual effects, but still fail to provide a satisfactory explanation in terms of realistic neural structures and cortical mechanisms, since the connection scheme they propose consists only of interactions among neurons with overlapping input fields. Here we propose an extended generative model for visual scenes that includes spatial dependencies among different features. We derive a neurophysiologically realistic inference scheme under the constraint that neurons have direct access only to local image information. The scheme can be interpreted as a network in primary visual cortex where two neural populations are organized in different layers within orientation hypercolumns that are connected by local, short-range and long-range recurrent interactions. When trained with natural images, the model predicts a connectivity structure linking neurons with similar orientation preferences matching the typical patterns found for long-ranging horizontal axons and feedback projections in visual cortex. Subjected to contextual stimuli typically used in empirical studies, our model replicates several hallmark effects of contextual processing and predicts characteristic differences for surround modulation between the two model populations. In summary, our model provides a novel framework for contextual processing in the visual system proposing a well-defined functional role for horizontal axons and feedback projections.

Contextual Integration in Cortical and Convolutional Neural Networks

“…Garrigues and Olshausen (2008) achieve this by including a pairwise coupling term in the prior for the sparse coding model. A recent study (Capparelli et al, 2019) achieves this by explicitly including spatial dependencies among dictionary elements with non-overlapping RFs into the sparse coding framework.…”

Prediction of future input explains lateral connectivity in primary visual cortex

“…Sparse coding argues that the brain is optimized to represent stimuli efficiently such that only a small number of neurons are strongly activated at a given time. When trained on static images, sparse coding models have been shown to replicate the like-for-like connectivity pattern among units with similar orientation tuning 40 . Where motion has been included in these models, they have also been shown to capture the asymmetry in excitatory and inhibitory inputs for direction tuning 41 .…”

Target Recognition of Industrial Robots Using Machine Vision in 5G Environment

“…It overcomes the information loss caused by multiple convolutions in the high-level convolution network, which makes the small target information in the highlevel features be seriously missed and the detection effects on small targets are poor. Meanwhile, due to the addition of the underlying features, the bottom layer features have rich direction information, which can help the angle prediction (Capparelli et al, 2019;Togaçar et al, 2020). Based on the original VGG-16 architecture, the 3rd, 4th, and 5th layers of the VGG-16 convolution layer are fused so that the model can have good detection robustness of smaller bottle objects on the conveyor belt; meanwhile, the information on the angle prediction of the object is increased by introducing a low-level neural response, thereby reducing the error of the angle prediction.…”