Meeting Agenda

Date:	January 10th, 2007
Location:	DC 1304
Time:	10:30
Chair:	Curtis Luk

1. Changes to the Agenda - additions or deletions

2. Coffee Hour

Coffee hour last week:: Robin- thanks!
Coffee hour this week:: Craig
Coffee hour next week:: Curtis

3. Forthcoming

Date:	January 17th	January 23rd	January 30th	February 6th
Location:	DC 1304 10:30	DC 1304 10:30	DC 1304 10:30	DC 1304 10:30
Chair:	Stephen Mann	Kevin Moule	Zheng Qin	Jamie Ruiz
Technical Presentation:	Robin Liu	Curtis Luk	Stephen Mann	Kevin Moule

4. Technical Presentation

Yi Lin	Title : Title: Analysis of Multi-Channel Time Series Using Anisotropic Diffusion Abstract: Time series representations are ubiquitous in information retrieval applications. The research on time series retrieval is becoming increasingly popular because its wide use in various applications, such as financial data analysis, weather forecast, music retrieval and motion synthesis for animations and games. One convenient way to investigate the data is to use existing examples as queries to find similar data, which is called content-based retrieval. However, efficient content-based retrieval in a large time series database depends on appropriate indexing and retrieval methods, which is still a challenging problem. In this talk, I will introduce my recent work on extracting features using anisotropic diffusion and use these features to index large time series databases. We segment the time series in a scale space after anisotropic diffusion filtering. Each segment is represented by a symbol, which is a group of parameters. When a query of time series given, we extracts its symbol sequence, compare this sequence with the symbol sequences of the database. The comparisons can be done from the course to fine levels using dynamic sequence alignment methods, such as Smith-Waterman algorithm. By matching the symbol sequences, a set of candidate matches is found and this set of data is small enough to be loaded into the main memory at one time.

Yi Lin

Title : Title: Analysis of Multi-Channel Time Series Using Anisotropic Diffusion

Abstract:
Time series representations are ubiquitous in information retrieval applications. The research on time series retrieval is becoming increasingly popular because its wide use in various applications, such as financial data analysis, weather forecast, music retrieval and motion synthesis for animations and games. One convenient way to investigate the data is to use existing examples as queries to find similar data, which is called content-based retrieval. However, efficient content-based retrieval in a large time series database depends on appropriate indexing and retrieval methods, which is still a challenging problem.

In this talk, I will introduce my recent work on extracting features using anisotropic diffusion and use these features to index large time series databases. We segment the time series in a scale space after anisotropic diffusion filtering. Each segment is represented by a symbol, which is a group of parameters. When a query of time series given, we extracts its symbol sequence, compare this sequence with the symbol sequences of the database. The comparisons can be done from the course to fine levels using dynamic sequence alignment methods, such as Smith-Waterman algorithm. By matching the symbol sequences, a set of candidate matches is found and this set of data is small enough to be loaded into the main memory at one time.

5. Discussion Items

Change time of lab meeting

6. Action Items

7. Conferences and Special Journal Issues

8. Directors' Meeting

9. Seminars and Events

Wednesday, 10 January 2007, 2:00PM - Computer Science (Algorithms and Complexity Group), DC 1304: Note: Slightly Different Time Than Usual: Jianer Chen: -- Improved Algorithms for Path, Matching and Packing Problems

Also see other Math and CS postings.