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

Pretrained deep models hold their learnt knowledge in the form of modelparameters. These parameters act as "memory" for the trained models and helpthem generalize well on unseen data. However, in absence of training data, theutility of a trained model is merely limited to either inference or betterinitialization towards a target task. In this paper, we go further and extractsynthetic data by leveraging the learnt model parameters. We dub them "DataImpressions", which act as proxy to the training data and can be used torealize a variety of tasks. These are useful in scenarios where only thepretrained models are available and the training data is not shared (e.g., dueto privacy or sensitivity concerns). We show the applicability of dataimpressions in solving several computer vision tasks such as unsuperviseddomain adaptation, continual learning as well as knowledge distillation. Wealso study the adversarial robustness of lightweight models trained viaknowledge distillation using these data impressions. Further, we demonstratethe efficacy of data impressions in generating data-free Universal AdversarialPerturbations (UAPs) with better fooling rates. Extensive experiments performedon benchmark datasets demonstrate competitive performance achieved using dataimpressions in absence of original training data.

Mining Data Impressions from Deep Models as Substitute for the Unavailable Training Data