Harnessing the Symmetry of Convolutions for Systematic Generalisation

Jeff Mitchell; Jeffrey S Bowers

doi:10.1109/IJCNN48605.2020.9207183

Harnessing the Symmetry of Convolutions for Systematic Generalisation

Jeff Mitchell, Jeffrey S Bowers

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

4 Citations (Scopus)

162 Downloads (Pure)

Abstract

We argue that symmetry is an important consideration in addressing the problem of systematic generalisation and investigate two forms of symmetry relevant to symbolic processes. We implement this approach in terms of convolution and show that it can be used to achieve effective generalisation in a rule learning and a context free language task.

In the rule learning task, we find that symmetry allows us to learn rules that abstract away from the particular symbols that instantiate them, enabling generalisation from seen to unseen symbols. In the language task, symmetry allows us to impose a stack like architecture on the memory cells of a recurrent net, which permits generalisation from simple to more complex structures.

Original language	English
Title of host publication	Harnessing the Symmetry of Convolutions for Systematic Generalisation
Publisher	Institute of Electrical and Electronics Engineers (IEEE)
Number of pages	8
DOIs	https://doi.org/10.1109/IJCNN48605.2020.9207183
Publication status	Published - 24 Jul 2020
Event	International Joint Conference on Neural Networks 2020 - Glasgow, United Kingdom Duration: 19 Jul 2020 → 24 Jul 2020 https://wcci2020.org/

Conference

Conference	International Joint Conference on Neural Networks 2020
Abbreviated title	IJCNN 2020
Country/Territory	United Kingdom
City	Glasgow
Period	19/07/20 → 24/07/20
Internet address	https://wcci2020.org/

Research Groups and Themes

Brain and Behaviour

Keywords

connectionism and neural nets
symbolic and algebraic manipulation
convolution

Access to Document

10.1109/IJCNN48605.2020.9207183

Full-text PDF (author’s accepted manuscript)
This is the author accepted manuscript (AAM). The final published version (version of record) is available online via Institute of Electrical and Electronics Engineers at https://ieeexplore.ieee.org/document/9207183 . Please refer to any applicable terms of use of the publisher.
Accepted author manuscript, 211 KB

Handle.net

Persistent link

M and M
Bowers, J. S. (Principal Investigator)
1/09/17 → 31/08/22
Project: Research, Parent

Cite this

@inproceedings{398bf965c0374d89bee4968cb55be92d,

title = "Harnessing the Symmetry of Convolutions for Systematic Generalisation",

abstract = "We argue that symmetry is an important consideration in addressing the problem of systematic generalisation and investigate two forms of symmetry relevant to symbolic processes. We implement this approach in terms of convolution and show that it can be used to achieve effective generalisation in a rule learning and a context free language task. In the rule learning task, we find that symmetry allows us to learn rules that abstract away from the particular symbols that instantiate them, enabling generalisation from seen to unseen symbols. In the language task, symmetry allows us to impose a stack like architecture on the memory cells of a recurrent net, which permits generalisation from simple to more complex structures. ",

keywords = "connectionism and neural nets, symbolic and algebraic manipulation, convolution",

author = "Jeff Mitchell and Bowers, {Jeffrey S}",

year = "2020",

month = jul,

day = "24",

doi = "10.1109/IJCNN48605.2020.9207183",

language = "English",

booktitle = "Harnessing the Symmetry of Convolutions for Systematic Generalisation",

publisher = "Institute of Electrical and Electronics Engineers (IEEE)",

address = "United States",

note = "International Joint Conference on Neural Networks 2020, IJCNN 2020 ; Conference date: 19-07-2020 Through 24-07-2020",

url = "https://wcci2020.org/",

}

Mitchell, J & Bowers, JS 2020, Harnessing the Symmetry of Convolutions for Systematic Generalisation. in Harnessing the Symmetry of Convolutions for Systematic Generalisation. Institute of Electrical and Electronics Engineers (IEEE), International Joint Conference on Neural Networks 2020, Glasgow, United Kingdom, 19/07/20. https://doi.org/10.1109/IJCNN48605.2020.9207183

TY - GEN

T1 - Harnessing the Symmetry of Convolutions for Systematic Generalisation

AU - Mitchell, Jeff

AU - Bowers, Jeffrey S

PY - 2020/7/24

Y1 - 2020/7/24

N2 - We argue that symmetry is an important consideration in addressing the problem of systematic generalisation and investigate two forms of symmetry relevant to symbolic processes. We implement this approach in terms of convolution and show that it can be used to achieve effective generalisation in a rule learning and a context free language task. In the rule learning task, we find that symmetry allows us to learn rules that abstract away from the particular symbols that instantiate them, enabling generalisation from seen to unseen symbols. In the language task, symmetry allows us to impose a stack like architecture on the memory cells of a recurrent net, which permits generalisation from simple to more complex structures.

AB - We argue that symmetry is an important consideration in addressing the problem of systematic generalisation and investigate two forms of symmetry relevant to symbolic processes. We implement this approach in terms of convolution and show that it can be used to achieve effective generalisation in a rule learning and a context free language task. In the rule learning task, we find that symmetry allows us to learn rules that abstract away from the particular symbols that instantiate them, enabling generalisation from seen to unseen symbols. In the language task, symmetry allows us to impose a stack like architecture on the memory cells of a recurrent net, which permits generalisation from simple to more complex structures.

KW - connectionism and neural nets

KW - symbolic and algebraic manipulation

KW - convolution

U2 - 10.1109/IJCNN48605.2020.9207183

DO - 10.1109/IJCNN48605.2020.9207183

M3 - Conference Contribution (Conference Proceeding)

BT - Harnessing the Symmetry of Convolutions for Systematic Generalisation

PB - Institute of Electrical and Electronics Engineers (IEEE)

T2 - International Joint Conference on Neural Networks 2020

Y2 - 19 July 2020 through 24 July 2020

ER -

Harnessing the Symmetry of Convolutions for Systematic Generalisation

Abstract

Conference

Research Groups and Themes

Keywords

Access to Document

Handle.net

Fingerprint

Projects

M and M

Cite this