ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models
With new accelerator hardware for DNN, the computing power for AI applications has increased rapidly. However, as DNN algorithms become more complex and optimized for specific applications, latency requirements remain challenging, and it is critical to find the optimal points in the design space. To...
Gespeichert in:
Veröffentlicht in: | arXiv.org 2021-05 |
---|---|
Hauptverfasser: | , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | |
---|---|
container_issue | |
container_start_page | |
container_title | arXiv.org |
container_volume | |
creator | Wess, Matthias Ivanov, Matvey Nookala, Anvesh Unger, Christoph Wendt, Alexander Jantsch, Axel |
description | With new accelerator hardware for DNN, the computing power for AI applications has increased rapidly. However, as DNN algorithms become more complex and optimized for specific applications, latency requirements remain challenging, and it is critical to find the optimal points in the design space. To decouple the architectural search from the target hardware, we propose a time estimation framework that allows for modeling the inference latency of DNNs on hardware accelerators based on mapping and layer-wise estimation models. The proposed methodology extracts a set of models from micro-kernel and multi-layer benchmarks and generates a stacked model for mapping and network execution time estimation. We compare estimation accuracy and fidelity of the generated mixed models, statistical models with the roofline model, and a refined roofline model for evaluation. We test the mixed models on the ZCU102 SoC board with DNNDK and Intel Neural Compute Stick 2 on a set of 12 state-of-the-art neural networks. It shows an average estimation error of 3.47% for the DNNDK and 7.44% for the NCS2, outperforming the statistical and analytical layer models for almost all selected networks. For a randomly selected subset of 34 networks of the NASBench dataset, the mixed model reaches fidelity of 0.988 in Spearman's rank correlation coefficient metric. The code of ANNETTE is publicly available at https://github.com/embedded-machine-learning/annette. |
doi_str_mv | 10.48550/arxiv.2105.03176 |
format | Article |
fullrecord | <record><control><sourceid>proquest_arxiv</sourceid><recordid>TN_cdi_arxiv_primary_2105_03176</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2524563453</sourcerecordid><originalsourceid>FETCH-LOGICAL-a523-b9d727885d2f631f3f84394003c385296289086b204adec8b44221c05fcd73ac3</originalsourceid><addsrcrecordid>eNotj1FLwzAURoMgOOZ-gE8GfO5M7k3a1LcyqhNmfbDvJU1T7NatM03d_PfWzqfDhY_LOYTccbYUSkr2qN25-V4CZ3LJkEfhFZkBIg-UALghi77fMsYgjEBKnJEsybI0z9MnmhgzOO0tzezIdoQ_dW5H07M1g2-6A82bvaVp75u9nu5T4z_ph9dmZyv61lW27W_Jda3b3i7-OSf5c5qv1sHm_eV1lWwCLQGDMq4iiJSSFdQh8hprJTAWjKFBJSEOQcVMhSUwoStrVClGd26YrE0VoTY4J_eXt1NscXSjkvsp_qKLKXpcPFwWR9d9Dbb3xbYb3GF0KkCCkCEKifgLIOlX3Q</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2524563453</pqid></control><display><type>article</type><title>ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models</title><source>arXiv.org</source><source>Free E- Journals</source><creator>Wess, Matthias ; Ivanov, Matvey ; Nookala, Anvesh ; Unger, Christoph ; Wendt, Alexander ; Jantsch, Axel</creator><creatorcontrib>Wess, Matthias ; Ivanov, Matvey ; Nookala, Anvesh ; Unger, Christoph ; Wendt, Alexander ; Jantsch, Axel</creatorcontrib><description>With new accelerator hardware for DNN, the computing power for AI applications has increased rapidly. However, as DNN algorithms become more complex and optimized for specific applications, latency requirements remain challenging, and it is critical to find the optimal points in the design space. To decouple the architectural search from the target hardware, we propose a time estimation framework that allows for modeling the inference latency of DNNs on hardware accelerators based on mapping and layer-wise estimation models. The proposed methodology extracts a set of models from micro-kernel and multi-layer benchmarks and generates a stacked model for mapping and network execution time estimation. We compare estimation accuracy and fidelity of the generated mixed models, statistical models with the roofline model, and a refined roofline model for evaluation. We test the mixed models on the ZCU102 SoC board with DNNDK and Intel Neural Compute Stick 2 on a set of 12 state-of-the-art neural networks. It shows an average estimation error of 3.47% for the DNNDK and 7.44% for the NCS2, outperforming the statistical and analytical layer models for almost all selected networks. For a randomly selected subset of 34 networks of the NASBench dataset, the mixed model reaches fidelity of 0.988 in Spearman's rank correlation coefficient metric. The code of ANNETTE is publicly available at https://github.com/embedded-machine-learning/annette.</description><identifier>EISSN: 2331-8422</identifier><identifier>DOI: 10.48550/arxiv.2105.03176</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Accelerators ; Accuracy ; Algorithms ; Computer Science - Hardware Architecture ; Computer Science - Learning ; Correlation coefficients ; Hardware ; Machine learning ; Mapping ; Multilayers ; Network latency ; Neural networks ; Statistical models</subject><ispartof>arXiv.org, 2021-05</ispartof><rights>2021. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,780,881,27902</link.rule.ids><backlink>$$Uhttps://doi.org/10.1109/ACCESS.2020.3047259$$DView published paper (Access to full text may be restricted)$$Hfree_for_read</backlink><backlink>$$Uhttps://doi.org/10.48550/arXiv.2105.03176$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Wess, Matthias</creatorcontrib><creatorcontrib>Ivanov, Matvey</creatorcontrib><creatorcontrib>Nookala, Anvesh</creatorcontrib><creatorcontrib>Unger, Christoph</creatorcontrib><creatorcontrib>Wendt, Alexander</creatorcontrib><creatorcontrib>Jantsch, Axel</creatorcontrib><title>ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models</title><title>arXiv.org</title><description>With new accelerator hardware for DNN, the computing power for AI applications has increased rapidly. However, as DNN algorithms become more complex and optimized for specific applications, latency requirements remain challenging, and it is critical to find the optimal points in the design space. To decouple the architectural search from the target hardware, we propose a time estimation framework that allows for modeling the inference latency of DNNs on hardware accelerators based on mapping and layer-wise estimation models. The proposed methodology extracts a set of models from micro-kernel and multi-layer benchmarks and generates a stacked model for mapping and network execution time estimation. We compare estimation accuracy and fidelity of the generated mixed models, statistical models with the roofline model, and a refined roofline model for evaluation. We test the mixed models on the ZCU102 SoC board with DNNDK and Intel Neural Compute Stick 2 on a set of 12 state-of-the-art neural networks. It shows an average estimation error of 3.47% for the DNNDK and 7.44% for the NCS2, outperforming the statistical and analytical layer models for almost all selected networks. For a randomly selected subset of 34 networks of the NASBench dataset, the mixed model reaches fidelity of 0.988 in Spearman's rank correlation coefficient metric. The code of ANNETTE is publicly available at https://github.com/embedded-machine-learning/annette.</description><subject>Accelerators</subject><subject>Accuracy</subject><subject>Algorithms</subject><subject>Computer Science - Hardware Architecture</subject><subject>Computer Science - Learning</subject><subject>Correlation coefficients</subject><subject>Hardware</subject><subject>Machine learning</subject><subject>Mapping</subject><subject>Multilayers</subject><subject>Network latency</subject><subject>Neural networks</subject><subject>Statistical models</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>BENPR</sourceid><sourceid>GOX</sourceid><recordid>eNotj1FLwzAURoMgOOZ-gE8GfO5M7k3a1LcyqhNmfbDvJU1T7NatM03d_PfWzqfDhY_LOYTccbYUSkr2qN25-V4CZ3LJkEfhFZkBIg-UALghi77fMsYgjEBKnJEsybI0z9MnmhgzOO0tzezIdoQ_dW5H07M1g2-6A82bvaVp75u9nu5T4z_ph9dmZyv61lW27W_Jda3b3i7-OSf5c5qv1sHm_eV1lWwCLQGDMq4iiJSSFdQh8hprJTAWjKFBJSEOQcVMhSUwoStrVClGd26YrE0VoTY4J_eXt1NscXSjkvsp_qKLKXpcPFwWR9d9Dbb3xbYb3GF0KkCCkCEKifgLIOlX3Q</recordid><startdate>20210507</startdate><enddate>20210507</enddate><creator>Wess, Matthias</creator><creator>Ivanov, Matvey</creator><creator>Nookala, Anvesh</creator><creator>Unger, Christoph</creator><creator>Wendt, Alexander</creator><creator>Jantsch, Axel</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20210507</creationdate><title>ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models</title><author>Wess, Matthias ; Ivanov, Matvey ; Nookala, Anvesh ; Unger, Christoph ; Wendt, Alexander ; Jantsch, Axel</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a523-b9d727885d2f631f3f84394003c385296289086b204adec8b44221c05fcd73ac3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Accelerators</topic><topic>Accuracy</topic><topic>Algorithms</topic><topic>Computer Science - Hardware Architecture</topic><topic>Computer Science - Learning</topic><topic>Correlation coefficients</topic><topic>Hardware</topic><topic>Machine learning</topic><topic>Mapping</topic><topic>Multilayers</topic><topic>Network latency</topic><topic>Neural networks</topic><topic>Statistical models</topic><toplevel>online_resources</toplevel><creatorcontrib>Wess, Matthias</creatorcontrib><creatorcontrib>Ivanov, Matvey</creatorcontrib><creatorcontrib>Nookala, Anvesh</creatorcontrib><creatorcontrib>Unger, Christoph</creatorcontrib><creatorcontrib>Wendt, Alexander</creatorcontrib><creatorcontrib>Jantsch, Axel</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science & Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection><collection>arXiv Computer Science</collection><collection>arXiv.org</collection><jtitle>arXiv.org</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Wess, Matthias</au><au>Ivanov, Matvey</au><au>Nookala, Anvesh</au><au>Unger, Christoph</au><au>Wendt, Alexander</au><au>Jantsch, Axel</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models</atitle><jtitle>arXiv.org</jtitle><date>2021-05-07</date><risdate>2021</risdate><eissn>2331-8422</eissn><abstract>With new accelerator hardware for DNN, the computing power for AI applications has increased rapidly. However, as DNN algorithms become more complex and optimized for specific applications, latency requirements remain challenging, and it is critical to find the optimal points in the design space. To decouple the architectural search from the target hardware, we propose a time estimation framework that allows for modeling the inference latency of DNNs on hardware accelerators based on mapping and layer-wise estimation models. The proposed methodology extracts a set of models from micro-kernel and multi-layer benchmarks and generates a stacked model for mapping and network execution time estimation. We compare estimation accuracy and fidelity of the generated mixed models, statistical models with the roofline model, and a refined roofline model for evaluation. We test the mixed models on the ZCU102 SoC board with DNNDK and Intel Neural Compute Stick 2 on a set of 12 state-of-the-art neural networks. It shows an average estimation error of 3.47% for the DNNDK and 7.44% for the NCS2, outperforming the statistical and analytical layer models for almost all selected networks. For a randomly selected subset of 34 networks of the NASBench dataset, the mixed model reaches fidelity of 0.988 in Spearman's rank correlation coefficient metric. The code of ANNETTE is publicly available at https://github.com/embedded-machine-learning/annette.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><doi>10.48550/arxiv.2105.03176</doi><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | EISSN: 2331-8422 |
ispartof | arXiv.org, 2021-05 |
issn | 2331-8422 |
language | eng |
recordid | cdi_arxiv_primary_2105_03176 |
source | arXiv.org; Free E- Journals |
subjects | Accelerators Accuracy Algorithms Computer Science - Hardware Architecture Computer Science - Learning Correlation coefficients Hardware Machine learning Mapping Multilayers Network latency Neural networks Statistical models |
title | ANNETTE: Accurate Neural Network Execution Time Estimation with Stacked Models |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-07T10%3A24%3A12IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_arxiv&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=ANNETTE:%20Accurate%20Neural%20Network%20Execution%20Time%20Estimation%20with%20Stacked%20Models&rft.jtitle=arXiv.org&rft.au=Wess,%20Matthias&rft.date=2021-05-07&rft.eissn=2331-8422&rft_id=info:doi/10.48550/arxiv.2105.03176&rft_dat=%3Cproquest_arxiv%3E2524563453%3C/proquest_arxiv%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2524563453&rft_id=info:pmid/&rfr_iscdi=true |