Compositional deep learning in Futhark
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
Compositional deep learning in Futhark. / Tran, Duc Minh; Henriksen, Troels; Elsman, Martin.
FHPNC 2019 - Proceedings of the 8th ACM SIGPLAN International Workshop on Functional High-Performance and Numerical Computing, co-located with ICFP 2019. ed. / Marco Zocca. Association for Computing Machinery, 2019. p. 47-59.Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - Compositional deep learning in Futhark
AU - Tran, Duc Minh
AU - Henriksen, Troels
AU - Elsman, Martin
PY - 2019/8/18
Y1 - 2019/8/18
N2 - We present a design pattern for composing deep learning networks in a typed, higher-order fashion. The exposed library functions are generically typed and the composition structure allows for networks to be trained (using backpropagation) and for trained networks to be used for predicting new results (using forward-propagation). Individual layers in a network can take different forms ranging over dense sigmoid layers to convolutional layers. The paper discusses different typing techniques aimed at enforcing proper use and composition of networks. The approach is implemented in Futhark, a data-parallel functional language and compiler targeting GPU architectures, and we demonstrate that Futhark's elimination of higher-order functions and modules leads to efficient generated code.
AB - We present a design pattern for composing deep learning networks in a typed, higher-order fashion. The exposed library functions are generically typed and the composition structure allows for networks to be trained (using backpropagation) and for trained networks to be used for predicting new results (using forward-propagation). Individual layers in a network can take different forms ranging over dense sigmoid layers to convolutional layers. The paper discusses different typing techniques aimed at enforcing proper use and composition of networks. The approach is implemented in Futhark, a data-parallel functional language and compiler targeting GPU architectures, and we demonstrate that Futhark's elimination of higher-order functions and modules leads to efficient generated code.
KW - Data-parallelism
KW - Deep learning
KW - Functional languages
UR - http://www.scopus.com/inward/record.url?scp=85072539228&partnerID=8YFLogxK
U2 - 10.1145/3331553.3342617
DO - 10.1145/3331553.3342617
M3 - Article in proceedings
SP - 47
EP - 59
BT - FHPNC 2019 - Proceedings of the 8th ACM SIGPLAN International Workshop on Functional High-Performance and Numerical Computing, co-located with ICFP 2019
A2 - Zocca, Marco
PB - Association for Computing Machinery
T2 - 8th ACM SIGPLAN International Workshop on Functional High-Performance and Numerical Computing, FHPNC 2019, co-located with ICFP 2019
Y2 - 18 August 2019
ER -
ID: 230447542