SSSL: Sequential Spatial and Structural Supervised Learning

This NSF-funded projects seeks to develop algorithms for learning to
classify items in sequential, spatial, and relational data. Application
projects include sequence labeling problems in bioinformatics (protein
secondary structure prediction, gene-finding, etc.), sequence labeling
problems in language processing (part-of-speech tagging, shallow
parsing, etc.), and pixel labeling problems in remote sensing (e.g.,
classifying pixels into land cover classes).

TaskTracer: Task-sensitive user interface for Windows

This NSF-managed project seeks to build a user interface that knows what
tasks you are currently working on and can help you carry out those
tasks. In particular, the system learns to predict your current task
and then provide easy access to relevant documents, email addresses, web
pages, and so on. We use the cluster to develop and test learning algorithms
for this project.

Knowledge-Intensive Learning

This DARPA and NSF-funded project has as its goal to bridge the gap
between knowledge representation and machine learning. Our goal is to
develop a system in which you can describe a learning problem in a
formal knowledge representation system and then the system automatically
formulates a learning system to solve that problem. This involves the
invention of features, selection among candidate features, and
extraction and learning with those features. Our application areas
include (a) modeling the spread of West Nile Virus and (b) predicting
grasshopper infestations in Eastern Oregon, and (c) learning for the
Task Tracer project.

INSECT-ID: Pattern Recognition of Insects for Environmental Modeling and Ecological Science

In this NSF-funded project, we are developing image processing and
learning algorithms for determining the genus and species of selected
classes of insects from image data. We are also constructing a
mechanical/optical device for manipulating and photographing insects.
We are using the cluster to perform the image processing and to develop and
test learning algorithms for this problem. Our two application tasks
are (a) measuring stream health by recognizing stone fly larvae in stream
substrate, and (b) measuring soil biodiversity by recognizing soil
mesofauna in forest soils.

Pedestrian Evacuation Modeling

As the war against terrorism escalates, office buildings, transportation facilities, and sports arenas become tempting targets. We are developing models of pedestrian motion and the spaces they occupy. We have a microscopic crowd evacuation simulator that moves each individual pedestrian separately. Our goal is to develop a
system capable of updating the positions of 100,000 or more people in real time.