HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency

HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency. Peng, B., Zhang, B., Chen, L., Avram, M., Henschel, R., Stewart, C., Zhu, S., McCallum, E., Smith, L., Zahniser, T., Omer, J., & Qiu, J. In Proceedings - 2017 IEEE International Conference on Big Data, Big Data 2017, volume 2018-Janua, pages 243-252, 2018. Institute of Electrical and Electronics Engineers Inc..

Website doi abstract bibtex

Latent Dirichlet Allocation (LDA) is a widely used machine learning technique in topic modeling and data analysis. Training large LDA models on big datasets involves dynamic and irregular computation patterns and is a major challenge to both algorithm optimization and system design. In this paper, we present a comprehensive benchmarking of our novel synchronized LDA training system HarpLDA+ based on Hadoop and Java. It demonstrates impressive performance when compared to three other MPI/C++ based state-of-the-art systems, which are LightLDA, F+NomadLDA, and WarpLDA. HarpLDA+ uses optimized collective communication with a timer control for load balance, leading to stable scalability in both shared-memory and distributed systems. We demonstrate in the experiments that HarpLDA+ is effective in reducing synchronization and communication overhead and outperforms the other three LDA training systems. © 2017 IEEE.

@inproceedings{
 title = {HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency},
 type = {inproceedings},
 year = {2018},
 keywords = {Algorithm optimization; Collective communications,Big data,Learning systems; Statistics},
 pages = {243-252},
 volume = {2018-Janua},
 websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85047792604&doi=10.1109%2FBigData.2017.8257932&partnerID=40&md5=634b9836d11831c51e661b9e5e90bcd9},
 publisher = {Institute of Electrical and Electronics Engineers Inc.},
 id = {ce97ebfa-0470-3e41-8c53-c8c9e7f45275},
 created = {2018-06-25T18:22:28.432Z},
 file_attached = {false},
 profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},
 last_modified = {2020-09-09T19:33:20.152Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {true},
 hidden = {false},
 citation_key = {Peng2018243},
 source_type = {conference},
 notes = {cited By 0; Conference of 5th IEEE International Conference on Big Data, Big Data 2017 ; Conference Date: 11 December 2017 Through 14 December 2017; Conference Code:134260},
 folder_uuids = {089a8687-5c2e-4a40-91e2-0a855ea1eb95},
 private_publication = {false},
 abstract = {Latent Dirichlet Allocation (LDA) is a widely used machine learning technique in topic modeling and data analysis. Training large LDA models on big datasets involves dynamic and irregular computation patterns and is a major challenge to both algorithm optimization and system design. In this paper, we present a comprehensive benchmarking of our novel synchronized LDA training system HarpLDA+ based on Hadoop and Java. It demonstrates impressive performance when compared to three other MPI/C++ based state-of-the-art systems, which are LightLDA, F+NomadLDA, and WarpLDA. HarpLDA+ uses optimized collective communication with a timer control for load balance, leading to stable scalability in both shared-memory and distributed systems. We demonstrate in the experiments that HarpLDA+ is effective in reducing synchronization and communication overhead and outperforms the other three LDA training systems. © 2017 IEEE.},
 bibtype = {inproceedings},
 author = {Peng, B and Zhang, B and Chen, L and Avram, M and Henschel, R and Stewart, C and Zhu, S and McCallum, E and Smith, L and Zahniser, T and Omer, J and Qiu, J},
 editor = {Obradovic Z. Baeza-Yates R., Kepner J Nambiar R Wang C Toyoda M Suzumura T Hu X Cuzzocrea A Baeza-Yates R Tang J Zang H Nie J.-Y. Ghosh R},
 doi = {10.1109/BigData.2017.8257932},
 booktitle = {Proceedings - 2017 IEEE International Conference on Big Data, Big Data 2017}
}

Downloads: 0

{"_id":"yHrX6Wh79uMoHQMoo","bibbaseid":"peng-zhang-chen-avram-henschel-stewart-zhu-mccallum-etal-harpldaoptimizinglatentdirichletallocationforparallelefficiency-2018","authorIDs":[],"author_short":["Peng, B.","Zhang, B.","Chen, L.","Avram, M.","Henschel, R.","Stewart, C.","Zhu, S.","McCallum, E.","Smith, L.","Zahniser, T.","Omer, J.","Qiu, J."],"bibdata":{"title":"HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency","type":"inproceedings","year":"2018","keywords":"Algorithm optimization; Collective communications,Big data,Learning systems; Statistics","pages":"243-252","volume":"2018-Janua","websites":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-85047792604&doi=10.1109%2FBigData.2017.8257932&partnerID=40&md5=634b9836d11831c51e661b9e5e90bcd9","publisher":"Institute of Electrical and Electronics Engineers Inc.","id":"ce97ebfa-0470-3e41-8c53-c8c9e7f45275","created":"2018-06-25T18:22:28.432Z","file_attached":false,"profile_id":"42d295c0-0737-38d6-8b43-508cab6ea85d","last_modified":"2020-09-09T19:33:20.152Z","read":false,"starred":false,"authored":"true","confirmed":"true","hidden":false,"citation_key":"Peng2018243","source_type":"conference","notes":"cited By 0; Conference of 5th IEEE International Conference on Big Data, Big Data 2017 ; Conference Date: 11 December 2017 Through 14 December 2017; Conference Code:134260","folder_uuids":"089a8687-5c2e-4a40-91e2-0a855ea1eb95","private_publication":false,"abstract":"Latent Dirichlet Allocation (LDA) is a widely used machine learning technique in topic modeling and data analysis. Training large LDA models on big datasets involves dynamic and irregular computation patterns and is a major challenge to both algorithm optimization and system design. In this paper, we present a comprehensive benchmarking of our novel synchronized LDA training system HarpLDA+ based on Hadoop and Java. It demonstrates impressive performance when compared to three other MPI/C++ based state-of-the-art systems, which are LightLDA, F+NomadLDA, and WarpLDA. HarpLDA+ uses optimized collective communication with a timer control for load balance, leading to stable scalability in both shared-memory and distributed systems. We demonstrate in the experiments that HarpLDA+ is effective in reducing synchronization and communication overhead and outperforms the other three LDA training systems. © 2017 IEEE.","bibtype":"inproceedings","author":"Peng, B and Zhang, B and Chen, L and Avram, M and Henschel, R and Stewart, C and Zhu, S and McCallum, E and Smith, L and Zahniser, T and Omer, J and Qiu, J","editor":"Obradovic Z. Baeza-Yates R., Kepner J Nambiar R Wang C Toyoda M Suzumura T Hu X Cuzzocrea A Baeza-Yates R Tang J Zang H Nie J.-Y. Ghosh R","doi":"10.1109/BigData.2017.8257932","booktitle":"Proceedings - 2017 IEEE International Conference on Big Data, Big Data 2017","bibtex":"@inproceedings{\n title = {HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency},\n type = {inproceedings},\n year = {2018},\n keywords = {Algorithm optimization; Collective communications,Big data,Learning systems; Statistics},\n pages = {243-252},\n volume = {2018-Janua},\n websites = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85047792604&doi=10.1109%2FBigData.2017.8257932&partnerID=40&md5=634b9836d11831c51e661b9e5e90bcd9},\n publisher = {Institute of Electrical and Electronics Engineers Inc.},\n id = {ce97ebfa-0470-3e41-8c53-c8c9e7f45275},\n created = {2018-06-25T18:22:28.432Z},\n file_attached = {false},\n profile_id = {42d295c0-0737-38d6-8b43-508cab6ea85d},\n last_modified = {2020-09-09T19:33:20.152Z},\n read = {false},\n starred = {false},\n authored = {true},\n confirmed = {true},\n hidden = {false},\n citation_key = {Peng2018243},\n source_type = {conference},\n notes = {cited By 0; Conference of 5th IEEE International Conference on Big Data, Big Data 2017 ; Conference Date: 11 December 2017 Through 14 December 2017; Conference Code:134260},\n folder_uuids = {089a8687-5c2e-4a40-91e2-0a855ea1eb95},\n private_publication = {false},\n abstract = {Latent Dirichlet Allocation (LDA) is a widely used machine learning technique in topic modeling and data analysis. Training large LDA models on big datasets involves dynamic and irregular computation patterns and is a major challenge to both algorithm optimization and system design. In this paper, we present a comprehensive benchmarking of our novel synchronized LDA training system HarpLDA+ based on Hadoop and Java. It demonstrates impressive performance when compared to three other MPI/C++ based state-of-the-art systems, which are LightLDA, F+NomadLDA, and WarpLDA. HarpLDA+ uses optimized collective communication with a timer control for load balance, leading to stable scalability in both shared-memory and distributed systems. We demonstrate in the experiments that HarpLDA+ is effective in reducing synchronization and communication overhead and outperforms the other three LDA training systems. © 2017 IEEE.},\n bibtype = {inproceedings},\n author = {Peng, B and Zhang, B and Chen, L and Avram, M and Henschel, R and Stewart, C and Zhu, S and McCallum, E and Smith, L and Zahniser, T and Omer, J and Qiu, J},\n editor = {Obradovic Z. Baeza-Yates R., Kepner J Nambiar R Wang C Toyoda M Suzumura T Hu X Cuzzocrea A Baeza-Yates R Tang J Zang H Nie J.-Y. Ghosh R},\n doi = {10.1109/BigData.2017.8257932},\n booktitle = {Proceedings - 2017 IEEE International Conference on Big Data, Big Data 2017}\n}","author_short":["Peng, B.","Zhang, B.","Chen, L.","Avram, M.","Henschel, R.","Stewart, C.","Zhu, S.","McCallum, E.","Smith, L.","Zahniser, T.","Omer, J.","Qiu, J."],"editor_short":["Obradovic Z. Baeza-Yates R., K., J., N., R., W., C., T., M., S., T., H., X., C., A., B., R., T., J., Z., H., N., J., G., R."],"urls":{"Website":"https://www.scopus.com/inward/record.uri?eid=2-s2.0-85047792604&doi=10.1109%2FBigData.2017.8257932&partnerID=40&md5=634b9836d11831c51e661b9e5e90bcd9"},"biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","bibbaseid":"peng-zhang-chen-avram-henschel-stewart-zhu-mccallum-etal-harpldaoptimizinglatentdirichletallocationforparallelefficiency-2018","role":"author","keyword":["Algorithm optimization; Collective communications","Big data","Learning systems; Statistics"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","creationDate":"2020-09-10T05:00:36.010Z","downloads":0,"keywords":["algorithm optimization; collective communications","big data","learning systems; statistics"],"search_terms":["harplda","optimizing","latent","dirichlet","allocation","parallel","efficiency","peng","zhang","chen","avram","henschel","stewart","zhu","mccallum","smith","zahniser","omer","qiu"],"title":"HarpLDA+: Optimizing latent dirichlet allocation for parallel efficiency","year":2018,"biburl":"https://bibbase.org/service/mendeley/42d295c0-0737-38d6-8b43-508cab6ea85d","dataSources":["zgahneP4uAjKbudrQ","ya2CyA73rpZseyrZ8","2252seNhipfTmjEBQ"]}