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

Binary neural networks (BNNs) have been widely adopted to reduce thecomputational cost and memory storage on edge-computing devices by usingone-bit representation for activations and weights. However, as neural networksbecome wider/deeper to improve accuracy and meet practical requirements, thecomputational burden remains a significant challenge even on the binaryversion. To address these issues, this paper proposes a novel method calledMinimum Spanning Tree (MST) compression that learns to compress and accelerateBNNs. The proposed architecture leverages an observation from previous worksthat an output channel in a binary convolution can be computed using anotheroutput channel and XNOR operations with weights that differ from the weights ofthe reused channel. We first construct a fully connected graph with verticescorresponding to output channels, where the distance between two vertices isthe number of different values between the weight sets used for these outputs.Then, the MST of the graph with the minimum depth is proposed to reorder outputcalculations, aiming to reduce computational cost and latency. Moreover, wepropose a new learning algorithm to reduce the total MST distance duringtraining. Experimental results on benchmark models demonstrate that our methodachieves significant compression ratios with negligible accuracy drops, makingit a promising approach for resource-constrained edge-computing devices.

MST-compression: Compressing and Accelerating Binary Neural Networks with Minimum Spanning Tree