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

This paper studies the problem of concept-based interpretability oftransformer representations for videos. Concretely, we seek to explain thedecision-making process of video transformers based on high-level,spatiotemporal concepts that are automatically discovered. Prior research onconcept-based interpretability has concentrated solely on image-level tasks.Comparatively, video models deal with the added temporal dimension, increasingcomplexity and posing challenges in identifying dynamic concepts over time. Inthis work, we systematically address these challenges by introducing the firstVideo Transformer Concept Discovery (VTCD) algorithm. To this end, we proposean efficient approach for unsupervised identification of units of videotransformer representations - concepts, and ranking their importance to theoutput of a model. The resulting concepts are highly interpretable, revealingspatio-temporal reasoning mechanisms and object-centric representations inunstructured video models. Performing this analysis jointly over a diverse setof supervised and self-supervised representations, we discover that some ofthese mechanism are universal in video transformers. Finally, we demonstratethat VTCDcan be used to improve model performance for fine-grained tasks.

Understanding Video Transformers via Universal Concept Discovery