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

Deep Learning (DL) has transformed the automation of a wide range ofindustries and finds increasing ubiquity in society. The high complexity of DLmodels and its widespread adoption has led to global energy consumptiondoubling every 3-4 months. Currently, the relationship between the DL modelconfiguration and energy consumption is not well established. At a generalcomputational energy model level, there is both strong dependency to both thehardware architecture (e.g. generic processors with different configuration ofinner components- CPU and GPU, programmable integrated circuits - FPGA), aswell as different interacting energy consumption aspects (e.g., data movement,calculation, control). At the DL model level, we need to translate non-linearactivation functions and its interaction with data into calculation tasks.Current methods mainly linearize nonlinear DL models to approximate itstheoretical FLOPs and MACs as a proxy for energy consumption. Yet, this isinaccurate (est. 93\% accuracy) due to the highly nonlinear nature of manyconvolutional neural networks (CNNs) for example.  In this paper, we develop a bottom-level Transistor Operations (TOs) methodto expose the role of non-linear activation functions and neural networkstructure in energy consumption. We translate a range of feedforward and CNNmodels into ALU calculation tasks and then TO steps. This is then statisticallylinked to real energy consumption values via a regression model for differenthardware configurations and data sets. We show that our proposed TOs method canachieve a 93.61% - 99.51% precision in predicting its energy consumption.

A Transistor Operations Model for Deep Learning Energy Consumption Scaling Law