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

Post-training quantization methods use a set of calibration data to computequantization ranges for network parameters and activations. The calibrationdata usually comes from the training dataset which could be inaccessible due tosensitivity of the data. In this work, we want to study such a problem: can weuse out-of-domain data to calibrate the trained networks without knowledge ofthe original dataset? Specifically, we go beyond the domain of natural imagesto include drastically different domains such as X-ray images, satellite imagesand ultrasound images. We find cross-domain calibration leads to surprisinglystable performance of quantized models on 10 tasks in different image domainswith 13 different calibration datasets. We also find that the performance ofquantized models is correlated with the similarity of the Gram matrices betweenthe source and calibration domains, which can be used as a criterion to choosecalibration set for better performance. We believe our research opens the doorto borrow cross-domain knowledge for network quantization and compression.

Is In-Domain Data Really Needed? A Pilot Study on Cross-Domain Calibration for Network Quantization