Efficient Representations for Lifelong Learning and Autoencoding

Maria-Florina Balcan, Avrim Blum, Santosh Vempala
; Proceedings of The 28th Conference on Learning Theory, PMLR 40:191-210, 2015.

Abstract

It has been a long-standing goal in machine learning, as well as in AI more generally, to develop life-long learning systems that learn many different tasks over time, and reuse insights from tasks learned, “learning to learn” as they do so. In this work we pose and provide efficient algorithms for several natural theoretical formulations of this goal. Specifically, we consider the problem of learning many different target functions over time, that share certain commonalities that are initially unknown to the learning algorithm. Our aim is to learn new internal representations as the algorithm learns new target functions, that capture this commonality and allow subsequent learning tasks to be solved more efficiently and from less data. We develop efficient algorithms for two very different kinds of commonalities that target functions might share: one based on learning common low-dimensional and unions of low-dimensional subspaces and one based on learning nonlinear Boolean combinations of features. Our algorithms for learning Boolean feature combinations additionally have a dual interpretation, and can be viewed as giving an efficient procedure for constructing near-optimal sparse Boolean autoencoders under a natural “anchor-set” assumption.

Cite this Paper


BibTeX
@InProceedings{pmlr-v40-Balcan15, title = {Efficient Representations for Lifelong Learning and Autoencoding}, author = {Maria-Florina Balcan and Avrim Blum and Santosh Vempala}, booktitle = {Proceedings of The 28th Conference on Learning Theory}, pages = {191--210}, year = {2015}, editor = {Peter Grünwald and Elad Hazan and Satyen Kale}, volume = {40}, series = {Proceedings of Machine Learning Research}, address = {Paris, France}, month = {03--06 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v40/Balcan15.pdf}, url = {http://proceedings.mlr.press/v40/Balcan15.html}, abstract = {It has been a long-standing goal in machine learning, as well as in AI more generally, to develop life-long learning systems that learn many different tasks over time, and reuse insights from tasks learned, “learning to learn” as they do so. In this work we pose and provide efficient algorithms for several natural theoretical formulations of this goal. Specifically, we consider the problem of learning many different target functions over time, that share certain commonalities that are initially unknown to the learning algorithm. Our aim is to learn new internal representations as the algorithm learns new target functions, that capture this commonality and allow subsequent learning tasks to be solved more efficiently and from less data. We develop efficient algorithms for two very different kinds of commonalities that target functions might share: one based on learning common low-dimensional and unions of low-dimensional subspaces and one based on learning nonlinear Boolean combinations of features. Our algorithms for learning Boolean feature combinations additionally have a dual interpretation, and can be viewed as giving an efficient procedure for constructing near-optimal sparse Boolean autoencoders under a natural “anchor-set” assumption.} }
Endnote
%0 Conference Paper %T Efficient Representations for Lifelong Learning and Autoencoding %A Maria-Florina Balcan %A Avrim Blum %A Santosh Vempala %B Proceedings of The 28th Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2015 %E Peter Grünwald %E Elad Hazan %E Satyen Kale %F pmlr-v40-Balcan15 %I PMLR %J Proceedings of Machine Learning Research %P 191--210 %U http://proceedings.mlr.press %V 40 %W PMLR %X It has been a long-standing goal in machine learning, as well as in AI more generally, to develop life-long learning systems that learn many different tasks over time, and reuse insights from tasks learned, “learning to learn” as they do so. In this work we pose and provide efficient algorithms for several natural theoretical formulations of this goal. Specifically, we consider the problem of learning many different target functions over time, that share certain commonalities that are initially unknown to the learning algorithm. Our aim is to learn new internal representations as the algorithm learns new target functions, that capture this commonality and allow subsequent learning tasks to be solved more efficiently and from less data. We develop efficient algorithms for two very different kinds of commonalities that target functions might share: one based on learning common low-dimensional and unions of low-dimensional subspaces and one based on learning nonlinear Boolean combinations of features. Our algorithms for learning Boolean feature combinations additionally have a dual interpretation, and can be viewed as giving an efficient procedure for constructing near-optimal sparse Boolean autoencoders under a natural “anchor-set” assumption.
RIS
TY - CPAPER TI - Efficient Representations for Lifelong Learning and Autoencoding AU - Maria-Florina Balcan AU - Avrim Blum AU - Santosh Vempala BT - Proceedings of The 28th Conference on Learning Theory PY - 2015/06/26 DA - 2015/06/26 ED - Peter Grünwald ED - Elad Hazan ED - Satyen Kale ID - pmlr-v40-Balcan15 PB - PMLR SP - 191 DP - PMLR EP - 210 L1 - http://proceedings.mlr.press/v40/Balcan15.pdf UR - http://proceedings.mlr.press/v40/Balcan15.html AB - It has been a long-standing goal in machine learning, as well as in AI more generally, to develop life-long learning systems that learn many different tasks over time, and reuse insights from tasks learned, “learning to learn” as they do so. In this work we pose and provide efficient algorithms for several natural theoretical formulations of this goal. Specifically, we consider the problem of learning many different target functions over time, that share certain commonalities that are initially unknown to the learning algorithm. Our aim is to learn new internal representations as the algorithm learns new target functions, that capture this commonality and allow subsequent learning tasks to be solved more efficiently and from less data. We develop efficient algorithms for two very different kinds of commonalities that target functions might share: one based on learning common low-dimensional and unions of low-dimensional subspaces and one based on learning nonlinear Boolean combinations of features. Our algorithms for learning Boolean feature combinations additionally have a dual interpretation, and can be viewed as giving an efficient procedure for constructing near-optimal sparse Boolean autoencoders under a natural “anchor-set” assumption. ER -
APA
Balcan, M., Blum, A. & Vempala, S.. (2015). Efficient Representations for Lifelong Learning and Autoencoding. Proceedings of The 28th Conference on Learning Theory, in PMLR 40:191-210

Related Material