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

Human ability to recognize complex visual patterns arises throughtransformations performed by successive areas in the ventral visual cortex.Deep neural networks trained end-to-end for object recognition approach humancapabilities, and offer the best descriptions to date of neural responses inthe late stages of the hierarchy. But these networks provide a poor account ofthe early stages, compared to traditional hand-engineered models, or modelsoptimized for coding efficiency or prediction. Moreover, the gradientbackpropagation used in end-to-end learning is generally considered to bebiologically implausible. Here, we overcome both of these limitations bydeveloping a bottom-up self-supervised training methodology that operatesindependently on successive layers. Specifically, we maximize featuresimilarity between pairs of locally-deformed natural image patches, whiledecorrelating features across patches sampled from other images. Crucially, thedeformation amplitudes are adjusted proportionally to receptive field sizes ineach layer, thus matching the task complexity to the capacity at each stage ofprocessing. In comparison with architecture-matched versions of previousmodels, we demonstrate that our layerwise complexity-matched learning (LCL)formulation produces a two-stage model (LCL-V2) that is better aligned withselectivity properties and neural activity in primate area V2. We demonstratethat the complexity-matched learning paradigm is critical for the emergence ofthe improved biological alignment. Finally, when the two-stage model is used asa fixed front-end for a deep network trained to perform object recognition, theresultant model (LCL-V2Net) is significantly better than standard end-to-endself-supervised, supervised, and adversarially-trained models in terms ofgeneralization to out-of-distribution tasks and alignment with human behavior.

Layerwise complexity-matched learning yields an improved model of cortical area V2