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

A data augmentation module is utilized in contrastive learning to transformthe given data example into two views, which is considered essential andirreplaceable. However, the predetermined composition of multiple dataaugmentations brings two drawbacks. First, the artificial choice ofaugmentation types brings specific representational invariances to the model,which have different degrees of positive and negative effects on differentdownstream tasks. Treating each type of augmentation equally during trainingmakes the model learn non-optimal representations for various downstream tasksand limits the flexibility to choose augmentation types beforehand. Second, thestrong data augmentations used in classic contrastive learning methods maybring too much invariance in some cases, and fine-grained information that isessential to some downstream tasks may be lost. This paper proposes a generalmethod to alleviate these two problems by considering where and what tocontrast in a general contrastive learning framework. We first propose to learndifferent augmentation invariances at different depths of the model accordingto the importance of each data augmentation instead of learningrepresentational invariances evenly in the backbone. We then propose to expandthe contrast content with augmentation embeddings to reduce the misleadingeffects of strong data augmentations. Experiments based on several baselinemethods demonstrate that we learn better representations for various benchmarkson classification, detection, and segmentation downstream tasks.

Rethinking the Augmentation Module in Contrastive Learning: Learning Hierarchical Augmentation Invariance with Expanded Views