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

Recently, an extensive amount of research has been focused on compressing andaccelerating Deep Neural Networks (DNN). So far, high compression ratealgorithms require part of the training dataset for a low precisioncalibration, or a fine-tuning process. However, this requirement isunacceptable when the data is unavailable or contains sensitive information, asin medical and biometric use-cases. We present three methods for generatingsynthetic samples from trained models. Then, we demonstrate how these samplescan be used to calibrate and fine-tune quantized models without using any realdata in the process. Our best performing method has a negligible accuracydegradation compared to the original training set. This method, which leveragesintrinsic batch normalization layers' statistics of the trained model, can beused to evaluate data similarity. Our approach opens a path towards genuinedata-free model compression, alleviating the need for training data duringmodel deployment.

The Knowledge Within: Methods for Data-Free Model Compression