### Abstract

An anomaly detection problem is investigated, in which s out of n sequences are anomalous and need to be detected. Each sequence consists of m independent and identically distributed (i.i.d.) samples drawn either from a nominal distribution p or from an anomalous distribution q that is distinct from p. Neither p nor q is known a priori. Two scenarios respectively with s known and unknown are studied. Distribution-free tests are constructed based on the metric of the maximum mean discrepancy (MMD). It is shown that if the value of s is known, as n goes to infinity, the number m of samples in each sequence should be of order O(log n) or larger to guarantee that the constructed test is exponentially consistent. On the other hand, if the value of s is unknown, the number m of samples in each sequence should be of the order strictly greater than O(log n) to guarantee the constructed test is consistent. The computational complexity of all tests are shown to be polynomial. Numerical results are provided to confirm the theoretic characterization of the performance. Further numerical results on both synthetic data sets and real data sets demonstrate that the MMD-based tests outperform or perform as well as other approaches.

Original language | English (US) |
---|---|

Title of host publication | 2014 52nd Annual Allerton Conference on Communication, Control, and Computing, Allerton 2014 |

Publisher | Institute of Electrical and Electronics Engineers Inc. |

Pages | 836-841 |

Number of pages | 6 |

ISBN (Print) | 9781479980093 |

DOIs | |

State | Published - Jan 30 2014 |

Event | 2014 52nd Annual Allerton Conference on Communication, Control, and Computing, Allerton 2014 - Monticello, United States Duration: Sep 30 2014 → Oct 3 2014 |

### Other

Other | 2014 52nd Annual Allerton Conference on Communication, Control, and Computing, Allerton 2014 |
---|---|

Country | United States |

City | Monticello |

Period | 9/30/14 → 10/3/14 |

### ASJC Scopus subject areas

- Computer Networks and Communications
- Computer Science Applications

## Fingerprint Dive into the research topics of 'Unsupervised nonparametric anomaly detection: A kernel method'. Together they form a unique fingerprint.

## Cite this

*2014 52nd Annual Allerton Conference on Communication, Control, and Computing, Allerton 2014*(pp. 836-841). [7028541] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ALLERTON.2014.7028541