A unified framework for consistency of regularized loss minimizers
We characterize a family of regularized loss minimization problems that satisfy three properties: scaled uniform convergence, super-norm regularization, and norm-loss monotonicity. We show several theoretical guarantees within this framework, including loss consistency, norm consistency, sparsistenc...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Article |
Language: | en_US |
Published: |
Association for Computing Machinery (ACM)
2015
|
Online Access: | http://hdl.handle.net/1721.1/100447 https://orcid.org/0000-0003-0238-6384 https://orcid.org/0000-0002-2199-0379 |
_version_ | 1811092489047638016 |
---|---|
author | Honorio, Jean Jaakkola, Tommi S. |
author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Honorio, Jean Jaakkola, Tommi S. |
author_sort | Honorio, Jean |
collection | MIT |
description | We characterize a family of regularized loss minimization problems that satisfy three properties: scaled uniform convergence, super-norm regularization, and norm-loss monotonicity. We show several theoretical guarantees within this framework, including loss consistency, norm consistency, sparsistency (i.e. support recovery) as well as sign consistency. A number of regularization problems can be shown to fall within our framework and we provide several examples. Our results can be seen as a concise summary of existing guarantees but we also extend them to new settings. Our formulation enables us to assume very little about the hypothesis class, data distribution, the loss, or the regularization. In particular, many of our results do not require a bounded hypothesis class, or identically distributed samples. Similarly, we do not assume boundedness, convexity or smoothness of the loss nor the regularizer. We only assume approximate optimality of the empirical minimizer. In terms of recovery, in contrast to existing results, our sparsistency and sign consistency results do not require knowledge of the sub-differential of the objective function. |
first_indexed | 2024-09-23T15:18:59Z |
format | Article |
id | mit-1721.1/100447 |
institution | Massachusetts Institute of Technology |
language | en_US |
last_indexed | 2024-09-23T15:18:59Z |
publishDate | 2015 |
publisher | Association for Computing Machinery (ACM) |
record_format | dspace |
spelling | mit-1721.1/1004472022-10-02T02:08:13Z A unified framework for consistency of regularized loss minimizers Honorio, Jean Jaakkola, Tommi S. Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Honorio, Jean Jaakkola, Tommi S. We characterize a family of regularized loss minimization problems that satisfy three properties: scaled uniform convergence, super-norm regularization, and norm-loss monotonicity. We show several theoretical guarantees within this framework, including loss consistency, norm consistency, sparsistency (i.e. support recovery) as well as sign consistency. A number of regularization problems can be shown to fall within our framework and we provide several examples. Our results can be seen as a concise summary of existing guarantees but we also extend them to new settings. Our formulation enables us to assume very little about the hypothesis class, data distribution, the loss, or the regularization. In particular, many of our results do not require a bounded hypothesis class, or identically distributed samples. Similarly, we do not assume boundedness, convexity or smoothness of the loss nor the regularizer. We only assume approximate optimality of the empirical minimizer. In terms of recovery, in contrast to existing results, our sparsistency and sign consistency results do not require knowledge of the sub-differential of the objective function. 2015-12-21T14:09:25Z 2015-12-21T14:09:25Z 2014 Article http://purl.org/eprint/type/ConferencePaper 1938-7228 http://hdl.handle.net/1721.1/100447 Honorio, Jean, and Tommi Jaakkola. "A unified framework for consistency of regularized loss minimizers." Journal of Machine Learning Research: Workshop and Conference Proceedings, Proceedings of The 31st International Conference on Machine Learning, Volume 32 (2014), 136-144. https://orcid.org/0000-0003-0238-6384 https://orcid.org/0000-0002-2199-0379 en_US http://jmlr.org/proceedings/papers/v32/ Journal of Machine Learning Research: Workshop and Conference Proceedings Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Association for Computing Machinery (ACM) MIT web domain |
spellingShingle | Honorio, Jean Jaakkola, Tommi S. A unified framework for consistency of regularized loss minimizers |
title | A unified framework for consistency of regularized loss minimizers |
title_full | A unified framework for consistency of regularized loss minimizers |
title_fullStr | A unified framework for consistency of regularized loss minimizers |
title_full_unstemmed | A unified framework for consistency of regularized loss minimizers |
title_short | A unified framework for consistency of regularized loss minimizers |
title_sort | unified framework for consistency of regularized loss minimizers |
url | http://hdl.handle.net/1721.1/100447 https://orcid.org/0000-0003-0238-6384 https://orcid.org/0000-0002-2199-0379 |
work_keys_str_mv | AT honoriojean aunifiedframeworkforconsistencyofregularizedlossminimizers AT jaakkolatommis aunifiedframeworkforconsistencyofregularizedlossminimizers AT honoriojean unifiedframeworkforconsistencyofregularizedlossminimizers AT jaakkolatommis unifiedframeworkforconsistencyofregularizedlossminimizers |