Modular meta-learning

Ferran Alet, Tomas Lozano-Perez, Leslie P. Kaelbling
Proceedings of The 2nd Conference on Robot Learning, PMLR 87:856-868, 2018.

Abstract

Many prediction problems, such as those that arise in the context of robotics, have a simplifying underlying structure that, if known, could accelerate learning. In this paper, we present a strategy for learning a set of neural network modules that can be combined in different ways. We train different modular structures on a set of related tasks and generalize to new tasks by composing the learned modules in new ways. By reusing modules to generalize we achieve combinatorial generalization, akin to the ”infinite use of finite means” displayed in language. Finally, we show this improves performance in two robotics-related problems.

Cite this Paper

BibTeX
@InProceedings{pmlr-v87-alet18a, title = {Modular meta-learning}, author = {Alet, Ferran and Lozano-Perez, Tomas and Kaelbling, Leslie P.}, booktitle = {Proceedings of The 2nd Conference on Robot Learning}, pages = {856--868}, year = {2018}, editor = {Billard, Aude and Dragan, Anca and Peters, Jan and Morimoto, Jun}, volume = {87}, series = {Proceedings of Machine Learning Research}, month = {29--31 Oct}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v87/alet18a/alet18a.pdf}, url = {https://proceedings.mlr.press/v87/alet18a.html}, abstract = {Many prediction problems, such as those that arise in the context of robotics, have a simplifying underlying structure that, if known, could accelerate learning. In this paper, we present a strategy for learning a set of neural network modules that can be combined in different ways. We train different modular structures on a set of related tasks and generalize to new tasks by composing the learned modules in new ways. By reusing modules to generalize we achieve combinatorial generalization, akin to the ”infinite use of finite means” displayed in language. Finally, we show this improves performance in two robotics-related problems.} }
Endnote
%0 Conference Paper %T Modular meta-learning %A Ferran Alet %A Tomas Lozano-Perez %A Leslie P. Kaelbling %B Proceedings of The 2nd Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2018 %E Aude Billard %E Anca Dragan %E Jan Peters %E Jun Morimoto %F pmlr-v87-alet18a %I PMLR %P 856--868 %U https://proceedings.mlr.press/v87/alet18a.html %V 87 %X Many prediction problems, such as those that arise in the context of robotics, have a simplifying underlying structure that, if known, could accelerate learning. In this paper, we present a strategy for learning a set of neural network modules that can be combined in different ways. We train different modular structures on a set of related tasks and generalize to new tasks by composing the learned modules in new ways. By reusing modules to generalize we achieve combinatorial generalization, akin to the ”infinite use of finite means” displayed in language. Finally, we show this improves performance in two robotics-related problems.
APA
Alet, F., Lozano-Perez, T. & Kaelbling, L.P.. (2018). Modular meta-learning. Proceedings of The 2nd Conference on Robot Learning, in Proceedings of Machine Learning Research 87:856-868 Available from https://proceedings.mlr.press/v87/alet18a.html.

Related Material