Berkeley Lab Scientific Computing Seminar

Date:

Wednesday, November 7, 2007

Time:

11:00am-12:00pm

Location:

Building 50A, 5132 Conference Room

Seminar Speaker:

Morten Mørup
Department of Intelligent Signal Processing
Technical University of Denmark, Denmark

Title:

Shift Invariant Data Decomposition

Abstract:

Data mining by decomposition based on low rank approximations have become an important tool in the analysis of many types of data. However, often such low rank approximations are too restrictive for the data at hand. One cause violating the simple low rank structure is the occurrence of shifts for instance caused by delays between the various sources and sensors measuring the data. We formulate models that can take into account shift in the data and derive efficient algorithms to solve for these types of shift invariant representations. The approach is generalizes to data of more than two modalities (i.e., tensors) and applications within image and sound processing as well as analysis of neuroscience data such as EEG and fMRI are given.

Sponsor of Seminar:

Esmond G. Ng

Scientific Computing

Contact Esmond G. Ng EGNg@lbl.gov