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

Amongst a variety of approaches aimed at making the learning procedure ofneural networks more effective, the scientific community developed strategiesto order the examples according to their estimated complexity, to distilknowledge from larger networks, or to exploit the principles behind adversarialmachine learning. A different idea has been recently proposed, named FriendlyTraining, which consists in altering the input data by adding an automaticallyestimated perturbation, with the goal of facilitating the learning process of aneural classifier. The transformation progressively fades-out as long astraining proceeds, until it completely vanishes. In this work we revisit andextend this idea, introducing a radically different and novel approach inspiredby the effectiveness of neural generators in the context of Adversarial MachineLearning. We propose an auxiliary multi-layer network that is responsible ofaltering the input data to make them easier to be handled by the classifier atthe current stage of the training procedure. The auxiliary network is trainedjointly with the neural classifier, thus intrinsically increasing the 'depth'of the classifier, and it is expected to spot general regularities in the dataalteration process. The effect of the auxiliary network is progressivelyreduced up to the end of training, when it is fully dropped and the classifieris deployed for applications. We refer to this approach as Neural FriendlyTraining. An extended experimental procedure involving several datasets anddifferent neural architectures shows that Neural Friendly Training overcomesthe originally proposed Friendly Training technique, improving thegeneralization of the classifier, especially in the case of noisy data.

Being Friends Instead of Adversaries: Deep Networks Learn from Data Simplified by Other Networks