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

Structured pruning and quantization are promising approaches for reducing theinference time and memory footprint of neural networks. However, most existingmethods require the original training dataset to fine-tune the model. This notonly brings heavy resource consumption but also is not possible forapplications with sensitive or proprietary data due to privacy and securityconcerns. Therefore, a few data-free methods are proposed to address thisproblem, but they perform data-free pruning and quantization separately, whichdoes not explore the complementarity of pruning and quantization. In thispaper, we propose a novel framework named Unified Data-Free Compression(UDFC),which performs pruning and quantization simultaneously without any data andfine-tuning process. Specifically, UDFC starts with the assumption that thepartial information of a damaged(e.g., pruned or quantized) channel can bepreserved by a linear combination of other channels, and then derives thereconstruction form from the assumption to restore the information loss due tocompression. Finally, we formulate the reconstruction error between theoriginal network and its compressed network, and theoretically deduce theclosed-form solution. We evaluate the UDFC on the large-scale imageclassification task and obtain significant improvements over various networkarchitectures and compression methods. For example, we achieve a 20.54%accuracy improvement on ImageNet dataset compared to SOTA method with 30%pruning ratio and 6-bit quantization on ResNet-34.

Unified Data-Free Compression: Pruning and Quantization without Fine-Tuning