An Explicitly Relational Neural Network Architecture

Murray Shanahan, Kyriacos Nikiforou, Antonia Creswell, Christos Kaplanis, David Barrett, Marta Garnelo
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:8593-8603, 2020.

Abstract

With a view to bridging the gap between deep learning and symbolic AI, we present a novel end-to-end neural network architecture that learns to form propositional representations with an explicitly relational structure from raw pixel data. In order to evaluate and analyse the architecture, we introduce a family of simple visual relational reasoning tasks of varying complexity. We show that the proposed architecture, when pre-trained on a curriculum of such tasks, learns to generate reusable representations that better facilitate subsequent learning on previously unseen tasks when compared to a number of baseline architectures. The workings of a successfully trained model are visualised to shed some light on how the architecture functions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-shanahan20a, title = {An Explicitly Relational Neural Network Architecture}, author = {Shanahan, Murray and Nikiforou, Kyriacos and Creswell, Antonia and Kaplanis, Christos and Barrett, David and Garnelo, Marta}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {8593--8603}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/shanahan20a/shanahan20a.pdf}, url = {https://proceedings.mlr.press/v119/shanahan20a.html}, abstract = {With a view to bridging the gap between deep learning and symbolic AI, we present a novel end-to-end neural network architecture that learns to form propositional representations with an explicitly relational structure from raw pixel data. In order to evaluate and analyse the architecture, we introduce a family of simple visual relational reasoning tasks of varying complexity. We show that the proposed architecture, when pre-trained on a curriculum of such tasks, learns to generate reusable representations that better facilitate subsequent learning on previously unseen tasks when compared to a number of baseline architectures. The workings of a successfully trained model are visualised to shed some light on how the architecture functions.} }
Endnote
%0 Conference Paper %T An Explicitly Relational Neural Network Architecture %A Murray Shanahan %A Kyriacos Nikiforou %A Antonia Creswell %A Christos Kaplanis %A David Barrett %A Marta Garnelo %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-shanahan20a %I PMLR %P 8593--8603 %U https://proceedings.mlr.press/v119/shanahan20a.html %V 119 %X With a view to bridging the gap between deep learning and symbolic AI, we present a novel end-to-end neural network architecture that learns to form propositional representations with an explicitly relational structure from raw pixel data. In order to evaluate and analyse the architecture, we introduce a family of simple visual relational reasoning tasks of varying complexity. We show that the proposed architecture, when pre-trained on a curriculum of such tasks, learns to generate reusable representations that better facilitate subsequent learning on previously unseen tasks when compared to a number of baseline architectures. The workings of a successfully trained model are visualised to shed some light on how the architecture functions.
APA
Shanahan, M., Nikiforou, K., Creswell, A., Kaplanis, C., Barrett, D. & Garnelo, M.. (2020). An Explicitly Relational Neural Network Architecture. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:8593-8603 Available from https://proceedings.mlr.press/v119/shanahan20a.html.

Related Material