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

Data augmentation is a critical contributing factor to the success of deeplearning but heavily relies on prior domain knowledge which is not alwaysavailable. Recent works on automatic data augmentation learn a policy to form asequence of augmentation operations, which are still pre-defined and restrictedto limited options. In this paper, we show that a prior-free autonomous dataaugmentation's objective can be derived from a representation learningprinciple that aims to preserve the minimum sufficient information of thelabels. Given an example, the objective aims at creating a distant "hardpositive example" as the augmentation, while still preserving the originallabel. We then propose a practical surrogate to the objective that can beoptimized efficiently and integrated seamlessly into existing methods for abroad class of machine learning tasks, e.g., supervised, semi-supervised, andnoisy-label learning. Unlike previous works, our method does not requiretraining an extra generative model but instead leverages the intermediate layerrepresentations of the end-task model for generating data augmentations. Inexperiments, we show that our method consistently brings non-trivialimprovements to the three aforementioned learning tasks from both efficiencyand final performance, either or not combined with strong pre-definedaugmentations, e.g., on medical images when domain knowledge is unavailable andthe existing augmentation techniques perform poorly. Code is available at:https://github.com/kai-wen-yang/LPA3}{https://github.com/kai-wen-yang/LPA3.

Adversarial Auto-Augment with Label Preservation: A Representation Learning Principle Guided Approach