English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The abundant recurrent horizontal and feedback connections in the primatevisual cortex are thought to play an important role in bringing global andsemantic contextual information to early visual areas during perceptualinference, helping to resolve local ambiguity and fill in missing details. Inthis study, we find that introducing feedback loops and horizontal recurrentconnections to a deep convolution neural network (VGG16) allows the network tobecome more robust against noise and occlusion during inference, even in theinitial feedforward pass. This suggests that recurrent feedback and contextualmodulation transform the feedforward representations of the network in ameaningful and interesting way. We study the population codes of neurons in thenetwork, before and after learning with feedback, and find that learning withfeedback yielded an increase in discriminability (measured by d-prime) betweenthe different object classes in the population codes of the neurons in thefeedforward path, even at the earliest layer that receives feedback. We findthat recurrent feedback, by injecting top-down semantic meaning to thepopulation activities, helps the network learn better feedforward paths torobustly map noisy image patches to the latent representations corresponding toimportant visual concepts of each object class, resulting in greater robustnessof the network against noises and occlusion as well as better fine-grainedrecognition.

Recurrent Feedback Improves Feedforward Representations in Deep Neural Networks