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

Deep neural networks (DNNs) are increasingly deployed in differentapplications to achieve state-of-the-art performance. However, they are oftenapplied as a black box with limited understanding of what knowledge the modelhas learned from the data. In this paper, we focus on image classification andpropose a method to visualize and understand the class-wise knowledge(patterns) learned by DNNs under three different settings including natural,backdoor and adversarial. Different to existing visualization methods, ourmethod searches for a single predictive pattern in the pixel space to representthe knowledge learned by the model for each class. Based on the proposedmethod, we show that DNNs trained on natural (clean) data learn abstract shapesalong with some texture, and backdoored models learn a suspicious pattern forthe backdoored class. Interestingly, the phenomenon that DNNs can learn asingle predictive pattern for each class indicates that DNNs can learn abackdoor even from clean data, and the pattern itself is a backdoor trigger. Inthe adversarial setting, we show that adversarially trained models tend tolearn more simplified shape patterns. Our method can serve as a useful tool tobetter understand the knowledge learned by DNNs on different datasets underdifferent settings.

What Do Deep Nets Learn? Class-wise Patterns Revealed in the Input Space