## Abstract

The “projection method” is an approach to finding numerical approximations to the optimal filter for non linear stochastic filtering problems. One uses a Hilbert space structure on a space of probability densities to project the infinite dimensional stochastic differential equation given by the filtering problem onto a finite dimensional manifold inside the space of densities. This reduces the problem to finite dimensional stochastic differential equation.

Previously, the projection method has only been considered for the Hilbert space structure associated with the Hellinger metric. We show through the numerical example of the quadratic sensor that the approach also works well when one projects using the direct L 2 metric.

Previous implementations of projection methods have been limited to solving a single problem. We indicate how one can build a computational framework for applying the projection method more generally.

Previously, the projection method has only been considered for the Hilbert space structure associated with the Hellinger metric. We show through the numerical example of the quadratic sensor that the approach also works well when one projects using the direct L 2 metric.

Previous implementations of projection methods have been limited to solving a single problem. We indicate how one can build a computational framework for applying the projection method more generally.

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

Title of host publication | Geometric Science of Information |

Subtitle of host publication | First International Conference, GSI 2013, Paris, France, August 28-30, 2013: proceedings |

Editors | Frank Nielsen, Frederic Barbaresco |

Place of Publication | Heidelberg |

Publisher | Springer |

Pages | 685-692 |

Number of pages | 8 |

Volume | N/A |

Edition | N/A |

ISBN (Print) | 9783642400209 |

DOIs | |

Publication status | Published - 2013 |

### Publication series

Name | Lecture Notes in Computer Science |
---|---|

Publisher | Springer |

Volume | 8085 |