In robot map learning, sensor information must be combined with external information such as robot position and landmark locations to generate a spatial representation of a designated area. Noisy sensors and inaccurate encoders can complicate this task, propagating error through the learned information. Learning algorithms may also need to fuse spatial information provided by multiple sources, such as multiple independent scouting robots.
Most processes studied in computational biology are governed by spatial constraints. For example, the activity of a candidate drug molecule, the function of a protein (and hence of the gene that codes for it), the behaviors of viral RNA, and the transcription and translation processes that lead from gene to protein all are based on 3-dimensional molecular structure. Machine learning is being applied in the study of all these processes. While successful to some extent, these applications remain hampered by inadequate spatial representations, inability to explicitly reason about uncertainty regarding spatial information, and failure to merge different sources of spatial information.
By their very nature, machine learning algorithms that assist with computer vision tasks must reason about spatial information. Learning-based image segmentation and object recognition algorithms require identification of spatial features that provide the most discriminating visual cues. Application areas such as telecommunications, computational fluid dynamics, and geographic information systems share this common need to discover salient features in spatial data.
The goal of this workshop is to present recent research results on learning from spatial data, to discuss issues relevant to all researchers working in this area, and to generate new ideas and collaborations among participants.
Electronic versions of the final papers will be available on the workshop home page, and authors are encouraged to read the included papers prior to the meeting. A compiled set of papers will be distributed as working notes at the workshop.
This one-day workshop will consist of research presentations and a panel discussion. Presentations will cover diverse application areas, but will highlight issues relevant to all researchers focusing on spatial information. The panel discussion will be designed to stimulation discussion on core issues and generate ideas for continued work in this area.