Conference paper Open Access

A Novel Posit-based Fast Approximation of ELU Activation Function for Deep Neural Networks

Cococcioni, Marco; Rossi, Federico; Ruffaldi, Emanuele; Saponara, Sergio


MARC21 XML Export

<?xml version='1.0' encoding='UTF-8'?>
<record xmlns="http://www.loc.gov/MARC21/slim">
  <leader>00000nam##2200000uu#4500</leader>
  <controlfield tag="005">20210226130245.0</controlfield>
  <controlfield tag="001">4042854</controlfield>
  <datafield tag="711" ind1=" " ind2=" ">
    <subfield code="d">14/09/2022</subfield>
    <subfield code="a">SMARTCOMP2020</subfield>
    <subfield code="c">Virtual (Online)</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="u">University of Pisa</subfield>
    <subfield code="a">Rossi, Federico</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="u">MMI spa</subfield>
    <subfield code="a">Ruffaldi, Emanuele</subfield>
  </datafield>
  <datafield tag="700" ind1=" " ind2=" ">
    <subfield code="u">University of Pisa</subfield>
    <subfield code="a">Saponara, Sergio</subfield>
  </datafield>
  <datafield tag="856" ind1="4" ind2=" ">
    <subfield code="s">191429</subfield>
    <subfield code="z">md5:c3961c8b727d556a9817d356a46728d3</subfield>
    <subfield code="u">https://zenodo.org/record/4042854/files/ROSSI_C3_smart_computing_2020_su_3_pp.pdf</subfield>
  </datafield>
  <datafield tag="542" ind1=" " ind2=" ">
    <subfield code="l">open</subfield>
  </datafield>
  <datafield tag="856" ind1="4" ind2=" ">
    <subfield code="y">Conference website</subfield>
    <subfield code="u">https://www.smartcomp.org</subfield>
  </datafield>
  <datafield tag="260" ind1=" " ind2=" ">
    <subfield code="c">2020-09-22</subfield>
  </datafield>
  <datafield tag="909" ind1="C" ind2="O">
    <subfield code="p">openaire</subfield>
    <subfield code="o">oai:zenodo.org:4042854</subfield>
  </datafield>
  <datafield tag="909" ind1="C" ind2="4">
    <subfield code="v">2020 IEEE International Conference on Smart Computing (SMARTCOMP)</subfield>
    <subfield code="p">A Novel Posit-based Fast Approximation of ELU Activation Function for Deep Neural Networks</subfield>
  </datafield>
  <datafield tag="100" ind1=" " ind2=" ">
    <subfield code="u">University of Pisa</subfield>
    <subfield code="a">Cococcioni, Marco</subfield>
  </datafield>
  <datafield tag="245" ind1=" " ind2=" ">
    <subfield code="a">A Novel Posit-based Fast Approximation of ELU Activation Function for Deep Neural Networks</subfield>
  </datafield>
  <datafield tag="536" ind1=" " ind2=" ">
    <subfield code="c">826647</subfield>
    <subfield code="a">SGA1 (Specific Grant Agreement 1) OF THE EUROPEAN PROCESSOR INITIATIVE (EPI)</subfield>
  </datafield>
  <datafield tag="540" ind1=" " ind2=" ">
    <subfield code="u">https://creativecommons.org/licenses/by/1.0/legalcode</subfield>
    <subfield code="a">Creative Commons Attribution 1.0 Generic</subfield>
  </datafield>
  <datafield tag="650" ind1="1" ind2="7">
    <subfield code="a">cc-by</subfield>
    <subfield code="2">opendefinition.org</subfield>
  </datafield>
  <datafield tag="520" ind1=" " ind2=" ">
    <subfield code="a">&lt;p&gt;Nowadays, &amp;nbsp;real-time &amp;nbsp;applications &amp;nbsp;are &amp;nbsp;exploiting DNNs &amp;nbsp;more &amp;nbsp;and &amp;nbsp;more &amp;nbsp;for &amp;nbsp;computer &amp;nbsp;vision &amp;nbsp;and &amp;nbsp;image &amp;nbsp;recognition &amp;nbsp;tasks. &amp;nbsp;Such kind of applications are posing strict constraints in terms of both fast and efficient information representation and processing. New formats for representing real numbers have been proposed and among them the Posit format appears to be very promising, providing means &amp;nbsp;to &amp;nbsp;implement &amp;nbsp;fast &amp;nbsp;approximated &amp;nbsp;version &amp;nbsp;of widely &amp;nbsp;used activation functions in DNNs. Moreover, information processing performance &amp;nbsp;are &amp;nbsp;continuously &amp;nbsp;improved &amp;nbsp;thanks &amp;nbsp;to &amp;nbsp;advanced vectorized &amp;nbsp;SIMD &amp;nbsp;(single-instruction &amp;nbsp;multiple-data) &amp;nbsp;processor architectures &amp;nbsp;and &amp;nbsp;instructions &amp;nbsp;like &amp;nbsp;ARM &amp;nbsp;SVE (Scalable Vector Extension). This &amp;nbsp;paper explores both &amp;nbsp;approaches (Posit-based implementation of activation functions and vectorized SIMD processor architectures) to &amp;nbsp;obtain &amp;nbsp;faster &amp;nbsp;DNNs. &amp;nbsp;The &amp;nbsp;two &amp;nbsp;proposed &amp;nbsp;techniques &amp;nbsp;are able &amp;nbsp;to &amp;nbsp;speed &amp;nbsp;up &amp;nbsp;both &amp;nbsp;DNN training &amp;nbsp;and &amp;nbsp;inference steps.&amp;nbsp;&lt;/p&gt;</subfield>
  </datafield>
  <datafield tag="773" ind1=" " ind2=" ">
    <subfield code="n">doi</subfield>
    <subfield code="i">isIdenticalTo</subfield>
    <subfield code="a">10.1109/SMARTCOMP50058.2020.00053</subfield>
  </datafield>
  <datafield tag="773" ind1=" " ind2=" ">
    <subfield code="n">doi</subfield>
    <subfield code="i">isVersionOf</subfield>
    <subfield code="a">10.5281/zenodo.4042853</subfield>
  </datafield>
  <datafield tag="024" ind1=" " ind2=" ">
    <subfield code="a">10.5281/zenodo.4042854</subfield>
    <subfield code="2">doi</subfield>
  </datafield>
  <datafield tag="980" ind1=" " ind2=" ">
    <subfield code="a">publication</subfield>
    <subfield code="b">conferencepaper</subfield>
  </datafield>
</record>
72
150
views
downloads
All versions This version
Views 7272
Downloads 150150
Data volume 28.7 MB28.7 MB
Unique views 6262
Unique downloads 142142

Share

Cite as