View in EDS HTML Full Text PDF Full Text

Distributed Newton Methods for Deep Neural Networks.

Saved in:
Bibliographic Details
Title: Distributed Newton Methods for Deep Neural Networks.
Authors: Wang, Chien-Chih, Tan, Kent Loong, Chen, Chun-Ting, Lin, Yu-Hsiang, Keerthi, S. Sathiya, Mahajan, Dhruv, Sundararajan, S., Lin, Chih-Jen
Source: Neural Computation. Jun2018, Vol. 30 Issue 6, p1673-1724. 52p. 2 Diagrams, 5 Charts, 2 Graphs.
Subjects: Deep learning, Mathematical optimization, Machine learning, Artificial neural networks, Newton-Raphson method
Abstract: Deep learning involves a difficult nonconvex optimization problem with a large number of weights between any two adjacent layers of a deep structure. To handle large data sets or complicated networks, distributed training is needed, but the calculation of function, gradient, and Hessian is expensive. In particular, the communication and the synchronization cost may become a bottleneck. In this letter, we focus on situations where the model is distributedly stored and propose a novel distributed Newton method for training deep neural networks. By variable and feature-wise data partitions and some careful designs, we are able to explicitly use the Jacobian matrix for matrix-vector products in the Newton method. Some techniques are incorporated to reduce the running time as well as memory consumption. First, to reduce the communication cost, we propose a diagonalization method such that an approximate Newton direction can be obtained without communication between machines. Second, we consider subsampled Gauss-Newton matrices for reducing the running time as well as the communication cost. Third, to reduce the synchronization cost, we terminate the process of finding an approximate Newton direction even though some nodes have not finished their tasks. Details of some implementation issues in distributed environments are thoroughly investigated. Experiments demonstrate that the proposed method is effective for the distributed training of deep neural networks. Compared with stochastic gradient methods, it is more robust and may give better test accuracy. [ABSTRACT FROM AUTHOR]
Copyright of Neural Computation is the property of MIT Press and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Psychology and Behavioral Sciences Collection
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
Text:
  Availability: 1
Header DbId: pbh
DbLabel: Psychology and Behavioral Sciences Collection
An: 129770349
AccessLevel: 6
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Distributed Newton Methods for Deep Neural Networks.
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Wang%2C+Chien-Chih%22">Wang, Chien-Chih</searchLink><br /><searchLink fieldCode="AR" term="%22Tan%2C+Kent+Loong%22">Tan, Kent Loong</searchLink><br /><searchLink fieldCode="AR" term="%22Chen%2C+Chun-Ting%22">Chen, Chun-Ting</searchLink><br /><searchLink fieldCode="AR" term="%22Lin%2C+Yu-Hsiang%22">Lin, Yu-Hsiang</searchLink><br /><searchLink fieldCode="AR" term="%22Keerthi%2C+S%2E+Sathiya%22">Keerthi, S. Sathiya</searchLink><br /><searchLink fieldCode="AR" term="%22Mahajan%2C+Dhruv%22">Mahajan, Dhruv</searchLink><br /><searchLink fieldCode="AR" term="%22Sundararajan%2C+S%2E%22">Sundararajan, S.</searchLink><br /><searchLink fieldCode="AR" term="%22Lin%2C+Chih-Jen%22">Lin, Chih-Jen</searchLink>
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="JN" term="%22Neural+Computation%22">Neural Computation</searchLink>. Jun2018, Vol. 30 Issue 6, p1673-1724. 52p. 2 Diagrams, 5 Charts, 2 Graphs.
– Name: Subject
  Label: Subjects
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Deep+learning%22">Deep learning</searchLink><br /><searchLink fieldCode="DE" term="%22Mathematical+optimization%22">Mathematical optimization</searchLink><br /><searchLink fieldCode="DE" term="%22Machine+learning%22">Machine learning</searchLink><br /><searchLink fieldCode="DE" term="%22Artificial+neural+networks%22">Artificial neural networks</searchLink><br /><searchLink fieldCode="DE" term="%22Newton-Raphson+method%22">Newton-Raphson method</searchLink>
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: Deep learning involves a difficult nonconvex optimization problem with a large number of weights between any two adjacent layers of a deep structure. To handle large data sets or complicated networks, distributed training is needed, but the calculation of function, gradient, and Hessian is expensive. In particular, the communication and the synchronization cost may become a bottleneck. In this letter, we focus on situations where the model is distributedly stored and propose a novel distributed Newton method for training deep neural networks. By variable and feature-wise data partitions and some careful designs, we are able to explicitly use the Jacobian matrix for matrix-vector products in the Newton method. Some techniques are incorporated to reduce the running time as well as memory consumption. First, to reduce the communication cost, we propose a diagonalization method such that an approximate Newton direction can be obtained without communication between machines. Second, we consider subsampled Gauss-Newton matrices for reducing the running time as well as the communication cost. Third, to reduce the synchronization cost, we terminate the process of finding an approximate Newton direction even though some nodes have not finished their tasks. Details of some implementation issues in distributed environments are thoroughly investigated. Experiments demonstrate that the proposed method is effective for the distributed training of deep neural networks. Compared with stochastic gradient methods, it is more robust and may give better test accuracy. [ABSTRACT FROM AUTHOR]
– Name: AbstractSuppliedCopyright
  Label:
  Group: Ab
  Data: <i>Copyright of Neural Computation is the property of MIT Press and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract.</i> (Copyright applies to all Abstracts.)
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=pbh&AN=129770349
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1162/neco_a_01088
    Languages:
      – Code: eng
        Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 52
        StartPage: 1673
    Subjects:
      – SubjectFull: Deep learning
        Type: general
      – SubjectFull: Mathematical optimization
        Type: general
      – SubjectFull: Machine learning
        Type: general
      – SubjectFull: Artificial neural networks
        Type: general
      – SubjectFull: Newton-Raphson method
        Type: general
    Titles:
      – TitleFull: Distributed Newton Methods for Deep Neural Networks.
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Wang, Chien-Chih
      – PersonEntity:
          Name:
            NameFull: Tan, Kent Loong
      – PersonEntity:
          Name:
            NameFull: Chen, Chun-Ting
      – PersonEntity:
          Name:
            NameFull: Lin, Yu-Hsiang
      – PersonEntity:
          Name:
            NameFull: Keerthi, S. Sathiya
      – PersonEntity:
          Name:
            NameFull: Mahajan, Dhruv
      – PersonEntity:
          Name:
            NameFull: Sundararajan, S.
      – PersonEntity:
          Name:
            NameFull: Lin, Chih-Jen
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 06
              Text: Jun2018
              Type: published
              Y: 2018
          Identifiers:
            – Type: issn-print
              Value: 08997667
          Numbering:
            – Type: volume
              Value: 30
            – Type: issue
              Value: 6
          Titles:
            – TitleFull: Neural Computation
              Type: main
ResultId 1