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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Large-scale deep neural networks (DNNs) are both compute and memoryintensive. As the size of DNNs continues to grow, it is critical to improve theenergy efficiency and performance while maintaining accuracy. For DNNs, themodel size is an important factor affecting performance, scalability and energyefficiency. Weight pruning achieves good compression ratios but suffers fromthree drawbacks: 1) the irregular network structure after pruning; 2) theincreased training complexity; and 3) the lack of rigorous guarantee ofcompression ratio and inference accuracy. To overcome these limitations, thispaper proposes CirCNN, a principled approach to represent weights and processneural networks using block-circulant matrices. CirCNN utilizes the FastFourier Transform (FFT)-based fast multiplication, simultaneously reducing thecomputational complexity (both in inference and training) from O(n2) toO(nlogn) and the storage complexity from O(n2) to O(n), with negligibleaccuracy loss. Compared to other approaches, CirCNN is distinct due to itsmathematical rigor: it can converge to the same effectiveness as DNNs withoutcompression. The CirCNN architecture, a universal DNN inference engine that canbe implemented on various hardware/software platforms with configurable networkarchitecture. To demonstrate the performance and energy efficiency, we testCirCNN in FPGA, ASIC and embedded processors. Our results show that CirCNNarchitecture achieves very high energy efficiency and performance with a smallhardware footprint. Based on the FPGA implementation and ASIC synthesisresults, CirCNN achieves 6-102X energy efficiency improvements compared withthe best state-of-the-art results.

CirCNN: Accelerating and Compressing Deep Neural Networks Using Block-CirculantWeight Matrices