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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Modern high-scoring models of vision in the brain score competition do notstem from Vision Transformers. However, in this paper, we provide evidenceagainst the unexpected trend of Vision Transformers (ViT) being notperceptually aligned with human visual representations by showing how adual-stream Transformer, a CrossViT$~\textit{a la}$ Chen et al. (2021), under ajoint rotationally-invariant and adversarial optimization procedure yields 2ndplace in the aggregate Brain-Score 2022 competition(Schrimpf et al., 2020b)averaged across all visual categories, and at the time of the competition held1st place for the highest explainable variance of area V4. In addition, ourcurrent Transformer-based model also achieves greater explainable variance forareas V4, IT and Behaviour than a biologically-inspired CNN (ResNet50) thatintegrates a frontal V1-like computation module (Dapello et al.,2020). Toassess the contribution of the optimization scheme with respect to the CrossViTarchitecture, we perform several additional experiments on differentlyoptimized CrossViT's regarding adversarial robustness, common corruptionbenchmarks, mid-ventral stimuli interpretation and feature inversion. Againstour initial expectations, our family of results provides tentative support foran $\textit{"All roads lead to Rome"}$ argument enforced via a jointoptimization rule even for non biologically-motivated models of vision such asVision Transformers. Code is available athttps://github.com/williamberrios/BrainScore-Transformers

Joint rotational invariance and adversarial training of a dual-stream Transformer yields state of the art Brain-Score for Area V4