Home   Profile    Research    Publications    Personal     Contact

Switching Linear Dynamic Systems (SLDS)

An SLDS is an intuitive model to describe the nonlinear dynamics using a set of switching linear dynamic systems (LDSs).

We develop new efficient and robust inference methods for SLDSs and SLDS extensions.
In our recent work, SLDSs were used to automatically decode the behaviors and communication messages of bees from their motions after the motion patterns are learned from the training data.
The goals of the works are to contribute to machine learning researches, as well as biological applications.
GeorgiaTech has developed many useful computational tools to automate the biology studies of animals and insects. See BioTracking page.

People

Sangmin Oh, Frank Dellaert, James M. Rehg.

Applications in biology

The overall research results are summarized and submitted to IJCV special issue on Vision for Learning and is under review.
Current draft of the paper can be obtained in a [Tech-report version]. More details and previous publications are enlisted below.
For research purposes, the used dataset are uploaded below along with the pointers to the
videos researchers can use to render their own lablineg results on top of them.
Feel free to use your own algorithm on the honey bee dance dataset, and let me know any interesting results.

Honey Bee Dance Data

See this page here to download the honey bee dance dataset.

Representative Publications

Learning and Inferring Motion Patterns using Parametric Segmental Switching Linear Dynamic Systems
Sang Min Oh, James M. Rehg, Tucker Balch, Frank Dellaert
Submitted to International Journal of Computer Vision (IJCV) Special Issue on Learning for Vision, under review
(Tech-report version available for download below)
Parameterized duration modeling for switching linear dynamic systems [pdf]
Sang Min Oh, James M. Rehg, Frank Dellaert
IEEE 2006 International Conference on Computer Vision and Pattern Recognition (CVPR 2006), NYC.
Learning and Inference in Parametric Switching Linear Dynamic Systems [pdf]
Sang Min Oh, James M. Rehg, Tucker Balch, Frank Dellaert
IEEE 2005 International Conference on Computer Vision (ICCV-2005), Beijing, China.
Data-Driven MCMC for Learning and Inference in Switching Linear Dynamic Systems [pdf]
Sang Min Oh, James M. Rehg, Tucker Balch, Frank Dellaert
Twentieth National Conference on Artificial Intelligence (AAAI-2005), Pittsburgh, U.S.A.
[Nomination for the Best paper award]

Tech-Reports