Sparse deep predictive coding captures contour integration capabilities of the early visual system

Abstract

Both neurophysiological and psychophysical experiments have pointed out the crucial role of recurrent and feedback connections to process context-dependent information in the early visual cortex. While numerous models have accounted for feedback effects at either neural or representational level, none of them were able to bind those two levels of analysis. Is it possible to describe feedback effects at both levels using the same model? We answer this question by combining Predictive Coding (PC) and Sparse Coding (SC) into a hierarchical and convolutional framework applied to realistic problems. In the Sparse Deep Predictive Coding (SDPC) model, the SC component models the internal recurrent processing within each layer, and the PC component describes the interactions between layers using feedforward and feedback connections. Here, we train a 2-layered SDPC on two different databases of images, and we interpret it as a model of the early visual system (V1 & V2). We first demonstrate that once the training has converged, SDPC exhibits oriented and localized receptive fields in V1 and more complex features in V2. Second, we analyze the effects of feedback on the neural organization beyond the classical receptive field of V1 neurons using interaction maps. These maps are similar to association fields and reflect the Gestalt principle of good continuation. We demonstrate that feedback signals reorganize interaction maps and modulate neural activity to promote contour integration. Third, we demonstrate at the representational level that the SDPC feedback connections are able to overcome noise in input images. Therefore, the SDPC captures the association field principle at the neural level which results in a better reconstruction of blurred images at the representational level.

A predictive processing model of episodic memory and time perception

“…The literature contains various neural implementations of predictive processing, including deep generative models [59][60][61]. and models that aim to replace backpropagation with local learning rules [62,63].…”

Inference via sparse coding in a hierarchical vision model

“…This practice gives the model an understanding of the underlying distribution of the sparse coding responses. Yet another appealing sparse coding approach is to learn both of the bases at the same time as done by Boutin et al. (2021) , Zeiler and Fergus (2010) .…”

Dynamical processing of orientation precision in the primary visual cortex

“…+ min acc (10) where max acc and min acc are respectively the maximum and minimum accuracies of the decoder, k the steepness and τ the time constant of the function.…”

“…In practical terms, given a low-precision image of a lion moving through the savannah, a predictive coding system that seeks to avoid the lion should (relatively to a highprecision image) rely more on its "lion" internal prediction than on its "savannah grass" sensory input to produce an optimal behaviour. Hence, given the mounting evidence that PC captures the behaviour of the visual system (8)(9)(10), it is physiologically likely -and mathematically required by PC -that the first level of the visual hierarchy, V1, possesses a mechanism which can account for the precision of sensory inputs.…”