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/

Zendy Plus

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

Zendy Tools

Zendy Open

Emergent hardwares can support mixed precision CNN models inference thatassign different bitwidths for different layers. Learning to find an optimalmixed precision model that can preserve accuracy and satisfy the specificconstraints on model size and computation is extremely challenge due to thedifficult in training a mixed precision model and the huge space of allpossible bit quantizations. In this paper, we propose a novel soft BarrierPenalty based NAS (BP-NAS) for mixed precision quantization, which ensures allthe searched models are inside the valid domain defined by the complexityconstraint, thus could return an optimal model under the given constraint byconducting search only one time. The proposed soft Barrier Penalty isdifferentiable and can impose very large losses to those models outside thevalid domain while almost no punishment for models inside the valid domain,thus constraining the search only in the feasible domain. In addition, adifferentiable Prob-1 regularizer is proposed to ensure learning with NAS isreasonable. A distribution reshaping training strategy is also used to maketraining more stable. BP-NAS sets new state of the arts on both classification(Cifar-10, ImageNet) and detection (COCO), surpassing all the efficient mixedprecision methods designed manually and automatically. Particularly, BP-NASachieves higher mAP (up to 2.7\% mAP improvement) together with lower bitcomputation cost compared with the existing best mixed precision model on COCOdetection.

Search What You Want: Barrier Panelty NAS for Mixed Precision Quantization