### Abstract

We propose a new online learning algorithm which provably approximates maximum margin classifiers with bias, where the margin is defined in terms of p-norm distance. Although learning of linear classifiers with bias can be reduced to learning of those without bias, the known reduction might lose the margin and slow down the convergence of online learning algorithms. Our algorithm, unlike previous online learning algorithms, implicitly uses a new reduction which preserves the margin and avoids such possible deficiencies. Our preliminary experiments show that our algorithm runs much faster than previous algorithms especially when the underlying linear classifier has large bias.

Original language | English |
---|---|

Pages | 69-80 |

Number of pages | 12 |

Publication status | Published - Dec 1 2008 |

Event | 21st Annual Conference on Learning Theory, COLT 2008 - Helsinki, Finland Duration: Jul 9 2008 → Jul 12 2008 |

### Other

Other | 21st Annual Conference on Learning Theory, COLT 2008 |
---|---|

Country | Finland |

City | Helsinki |

Period | 7/9/08 → 7/12/08 |

### Fingerprint

### All Science Journal Classification (ASJC) codes

- Education

### Cite this

*Online learning of approximate maximum p-norm margin classifiers with bias*. 69-80. Paper presented at 21st Annual Conference on Learning Theory, COLT 2008, Helsinki, Finland.

**Online learning of approximate maximum p-norm margin classifiers with bias.** / Ishibashi, Kosuke; Hatano, Kohei; Takeda, Masayuki.

Research output: Contribution to conference › Paper

}

TY - CONF

T1 - Online learning of approximate maximum p-norm margin classifiers with bias

AU - Ishibashi, Kosuke

AU - Hatano, Kohei

AU - Takeda, Masayuki

PY - 2008/12/1

Y1 - 2008/12/1

N2 - We propose a new online learning algorithm which provably approximates maximum margin classifiers with bias, where the margin is defined in terms of p-norm distance. Although learning of linear classifiers with bias can be reduced to learning of those without bias, the known reduction might lose the margin and slow down the convergence of online learning algorithms. Our algorithm, unlike previous online learning algorithms, implicitly uses a new reduction which preserves the margin and avoids such possible deficiencies. Our preliminary experiments show that our algorithm runs much faster than previous algorithms especially when the underlying linear classifier has large bias.

AB - We propose a new online learning algorithm which provably approximates maximum margin classifiers with bias, where the margin is defined in terms of p-norm distance. Although learning of linear classifiers with bias can be reduced to learning of those without bias, the known reduction might lose the margin and slow down the convergence of online learning algorithms. Our algorithm, unlike previous online learning algorithms, implicitly uses a new reduction which preserves the margin and avoids such possible deficiencies. Our preliminary experiments show that our algorithm runs much faster than previous algorithms especially when the underlying linear classifier has large bias.

UR - http://www.scopus.com/inward/record.url?scp=80054116173&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=80054116173&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:80054116173

SP - 69

EP - 80

ER -