GCOEセミナー
過去の記録 ~05/02|次回の予定|今後の予定 05/03~
2013年01月24日(木)
15:00-16:00 数理科学研究科棟(駒場) 056号室
Christian Clason 氏 (Graz University)
Parameter identification problems with non-Gaussian noise (ENGLISH)
Christian Clason 氏 (Graz University)
Parameter identification problems with non-Gaussian noise (ENGLISH)
[ 講演概要 ]
For inverse problems subject to non-Gaussian (such as impulsive or uniform) noise, other data fitting terms than the standard L^2 norm are statistically appropriate and more robust. However, these formulations typically lead to non-differentiable problems which are challenging to solve numerically. This talk presents an approach that combines an iterative smoothing procedure with a semismooth Newton method, which can be applied to parameter identification problems for partial differential equations. The efficiency of this approach is illustrated for the inverse potential problem.
For inverse problems subject to non-Gaussian (such as impulsive or uniform) noise, other data fitting terms than the standard L^2 norm are statistically appropriate and more robust. However, these formulations typically lead to non-differentiable problems which are challenging to solve numerically. This talk presents an approach that combines an iterative smoothing procedure with a semismooth Newton method, which can be applied to parameter identification problems for partial differential equations. The efficiency of this approach is illustrated for the inverse potential problem.