Grid-Functioned Neural Networks

Javier Dehesa; Andrew Vidler; Julian Padget; Christof Lutteroth

Grid-Functioned Neural Networks

Javier Dehesa, Andrew Vidler, Julian Padget, Christof Lutteroth

Ninja Theory Ltd

Research output: Chapter or section in a book/report/conference proceeding › Chapter in a published conference proceeding

3 Citations (SciVal)

80 Downloads (Pure)

Abstract

We introduce a new neural network architecture that we call "grid-functioned" neural networks. It utilises a grid structure of network parameterisations that can be specialised for different subdomains of the problem, while maintaining smooth, continuous behaviour. The grid gives the user flexibility to prevent gross features from overshadowing important minor ones. We present a full characterisation of its computational and spatial complexity, and demonstrate its potential, compared to a traditional architecture, over a set of synthetic regression problems. We further illustrate the benefits through a real-world 3D skeletal animation case study, where it offers the same visual quality as a state-of-the-art model, but with lower computational complexity and better control accuracy.

Original language	English
Title of host publication	Proceedings of the Thirty-eighth International Conference on Machine Learning (ICML 2021)
Publisher	Curran Associates, Inc.
Pages	2559-2567
Number of pages	9
Volume	2021
Edition	139
Publication status	Published - 18 Jul 2021
Event	Thirty-eighth International Conference on Machine Learning - Virutal only Duration: 18 Jul 2021 → 24 Jul 2021 Conference number: 38 https://icml.cc/Conferences/2021

Publication series

Name	Proceedings of Machine Learning Research
Publisher	PMLR
Volume	139
ISSN (Electronic)	2640-3498

Conference

Conference	Thirty-eighth International Conference on Machine Learning
Abbreviated title	ICML
Period	18/07/21 → 24/07/21
Internet address	https://icml.cc/Conferences/2021

Keywords

Machine Learning
Neural Networks

Access to Document

Grid-Functioned Neural NetworksAccepted author manuscript, 2.68 MB

http://proceedings.mlr.press/v139/dehesa21a.html

EPSRC Centre for Doctoral Training in Digital Entertainment
Willis, P. (PI)
1/04/14 → 30/09/22
Project: Research-related funding

Cite this

Grid-Functioned Neural Networks. / Dehesa, Javier; Vidler, Andrew; Padget, Julian et al.
Proceedings of the Thirty-eighth International Conference on Machine Learning (ICML 2021). Vol. 2021 139. ed. Curran Associates, Inc., 2021. p. 2559-2567 (Proceedings of Machine Learning Research; Vol. 139).

Research output: Chapter or section in a book/report/conference proceeding › Chapter in a published conference proceeding

Dehesa, J, Vidler, A, Padget, J & Lutteroth, C 2021, Grid-Functioned Neural Networks. in Proceedings of the Thirty-eighth International Conference on Machine Learning (ICML 2021). 139 edn, vol. 2021, Proceedings of Machine Learning Research, vol. 139, Curran Associates, Inc., pp. 2559-2567, Thirty-eighth International Conference on Machine Learning, 18/07/21. <http://proceedings.mlr.press/v139/dehesa21a.html>

@inproceedings{75e21b13dd6346faa29dfad7adad0b1e,

title = "Grid-Functioned Neural Networks",

abstract = "We introduce a new neural network architecture that we call {"}grid-functioned{"} neural networks. It utilises a grid structure of network parameterisations that can be specialised for different subdomains of the problem, while maintaining smooth, continuous behaviour. The grid gives the user flexibility to prevent gross features from overshadowing important minor ones. We present a full characterisation of its computational and spatial complexity, and demonstrate its potential, compared to a traditional architecture, over a set of synthetic regression problems. We further illustrate the benefits through a real-world 3D skeletal animation case study, where it offers the same visual quality as a state-of-the-art model, but with lower computational complexity and better control accuracy.",

keywords = "Machine Learning, Neural Networks",

author = "Javier Dehesa and Andrew Vidler and Julian Padget and Christof Lutteroth",

year = "2021",

month = jul,

day = "18",

language = "English",

volume = "2021",

series = "Proceedings of Machine Learning Research",

publisher = "Curran Associates, Inc.",

pages = "2559--2567",

booktitle = "Proceedings of the Thirty-eighth International Conference on Machine Learning (ICML 2021)",

address = "USA United States",

edition = "139",

note = "Thirty-eighth International Conference on Machine Learning, ICML ; Conference date: 18-07-2021 Through 24-07-2021",

url = "https://icml.cc/Conferences/2021",

}

TY - GEN

T1 - Grid-Functioned Neural Networks

AU - Dehesa, Javier

AU - Vidler, Andrew

AU - Padget, Julian

AU - Lutteroth, Christof

N1 - Conference code: 38

PY - 2021/7/18

Y1 - 2021/7/18

N2 - We introduce a new neural network architecture that we call "grid-functioned" neural networks. It utilises a grid structure of network parameterisations that can be specialised for different subdomains of the problem, while maintaining smooth, continuous behaviour. The grid gives the user flexibility to prevent gross features from overshadowing important minor ones. We present a full characterisation of its computational and spatial complexity, and demonstrate its potential, compared to a traditional architecture, over a set of synthetic regression problems. We further illustrate the benefits through a real-world 3D skeletal animation case study, where it offers the same visual quality as a state-of-the-art model, but with lower computational complexity and better control accuracy.

AB - We introduce a new neural network architecture that we call "grid-functioned" neural networks. It utilises a grid structure of network parameterisations that can be specialised for different subdomains of the problem, while maintaining smooth, continuous behaviour. The grid gives the user flexibility to prevent gross features from overshadowing important minor ones. We present a full characterisation of its computational and spatial complexity, and demonstrate its potential, compared to a traditional architecture, over a set of synthetic regression problems. We further illustrate the benefits through a real-world 3D skeletal animation case study, where it offers the same visual quality as a state-of-the-art model, but with lower computational complexity and better control accuracy.

KW - Machine Learning

KW - Neural Networks

M3 - Chapter in a published conference proceeding

VL - 2021

T3 - Proceedings of Machine Learning Research

SP - 2559

EP - 2567

BT - Proceedings of the Thirty-eighth International Conference on Machine Learning (ICML 2021)

PB - Curran Associates, Inc.

T2 - Thirty-eighth International Conference on Machine Learning

Y2 - 18 July 2021 through 24 July 2021

ER -

Grid-Functioned Neural Networks

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Projects

EPSRC Centre for Doctoral Training in Digital Entertainment

Cite this