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

In recent years, researchers have focused on reducing the model size andnumber of computations (measured as "multiply-accumulate" or MAC operations) ofDNNs. The energy consumption of a DNN depends on both the number of MACoperations and the energy efficiency of each MAC operation. The former can beestimated at design time; however, the latter depends on the intricate datareuse patterns and underlying hardware architecture. Hence, estimating it atdesign time is challenging. This work shows that the conventional approach toestimate the data reuse, viz. arithmetic intensity, does not always correctlyestimate the degree of data reuse in DNNs since it gives equal importance toall the data types. We propose a novel model, termed "data type aware weightedarithmetic intensity" ($DI$), which accounts for the unequal importance ofdifferent data types in DNNs. We evaluate our model on 25 state-of-the-art DNNson two GPUs. We show that our model accurately models data-reuse for allpossible data reuse patterns for different types of convolution and differenttypes of layers. We show that our model is a better indicator of the energyefficiency of DNNs. We also show its generality using the central limittheorem.

Modeling Data Reuse in Deep Neural Networks by Taking Data-Types into Cognizance