Effect of Top-Down Connections in Hierarchical Sparse Coding

Boutin, Victor; Franciosini, Angelo; Ruffier, Franck; Perrinet, Laurent

doi:10.1162/neco_a_01325

Cited by 14 publications

(18 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We thus call k FB 'the feedback strength' as it allows us to tune how close the V1 neural activity is from its prediction made by V2. Last but not least, when the parameter k FB is set to 0, the SDPC becomes a stacking of independent LASSO sub-problems [38,39] and is not relying anymore on the Predictive Coding (PC) framework. Consequently, we also use the k FB parameter to evaluate the effect of the PC on the first layer representation.…”

Section: Brief Description Of the Sdpcmentioning

confidence: 99%

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

et al. 2021

Self Cite

View full text Add to dashboard Cite

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.

show abstract

Section: Brief Description Of the Sdpcmentioning

confidence: 99%

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

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…A group of neurons γ i predicts, at best, the activity from the previous cortical layer γ i−1 , trough a set of synaptic weights W i . Given a network with N layers, we can define the generative model [23,24] as:…”

Section: The Sdpc Frameworkmentioning

confidence: 99%

“…The best γ i for each layer is inferred through a multi-layer version of the Fast Iterative Thresholding Algorithm (FISTA) [32], introduced in [24]. This process is usually referred to as inference [33].…”

Section: The Sdpc Frameworkmentioning

confidence: 99%

“…This was the case for the different network sizes (36,49,64, 81, 100 and 121 neurons for each layer) and the different combinations of pooling functions: MaxPool 2D S , MaxPool 2D F , MaxPool 2D S + 1D F , and MaxPool 2D S + 2D F . As in [23] and [24], the network efficiently developed edge-like receptive fields (RFs) in the first layer (see Fig. 1,b).…”

Section: Pooling In a Predictive Coding Networkmentioning

confidence: 99%

“…To take advantage of the modeling power of SC and PC we integrated both these mechanisms in a model of V1, called Sparse Deep Predictive Coding (SDPC). This framework introduces biologically realistic convolutional architecture which integrates predictive coding while accounting for sparsity constraints [23,24]. Yet, both these SC and PC models are missing a third key component necessary to model the emergence of complex cells: nonlinear pooling.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Pooling in a predictive model of V1 explains functional and structural diversity across species

Franciosini¹,

Boutin²,

Chavane³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Neurons in the primary visual cortex are selective to orientation with various degrees of selectivity to the spatial phase, from high selectivity in simple cells to low selectivity in complex cells. Various computational models have suggested a possible link between the presence of phase invariant cells and the existence of cortical orientation maps in higher mammals' V1. These models, however, do not explain the emergence of complex cells in animals that do not show orientation maps. In this study, we build a model of V1 based on a convolutional network called Sparse Deep Predictive Coding (SDPC) and show that a single computational mechanism, pooling, allows the SDPC model to account for the emergence of complex cells as well as cortical orientation maps in V1, as observed in distinct species of mammals. By using different pooling functions, our model developed complex cells in networks that exhibit orientation maps (e.g., like in carnivores and primates) or not (e.g., rodents and lagomorphs). The SDPC can therefore be viewed as a unifying framework that explains the diversity of structural and functional phenomena observed in V1. In particular, we show that orientation maps emerge naturally as the most cost-efficient structure to generate complex cells under the predictive coding principle.

show abstract