There are well-established methods for reducing the number of support vectors in a trained binary support vector machine, often with minimal impact on accuracy. We show how reduced-set methods can be applied to multiclass SVMs made up of several binary SVMs, with significantly better results than reducing each binary SVM independently. Our approach is based on Burges' approach that constructs each reduced-set vector as the pre-image of a vector in kernel space, but we extend this by recomputing the SVM weights and bias optimally using the original SVM objective function. This leads to greater accuracy for a binary reduced-set SVM, and also allows vectors to be "shared" between multiple binary SVMs for greater multiclass accuracy with fewer reduced-set vectors. We also propose computing pre-images using differential evolution, which we have found to be more robust than gradient descent alone. We show experimental results on a variety of problems and find that this new approach is consistently better than previous multiclass reduced-set methods, sometimes with a dramatic difference.

Typically, data collected by a spacecraft is downlinked to Earth and pre-processed before any analysis is performed. We have developed classifiers that can be used onboard a spacecraft to identify high priority data for downlink to Earth, providing a method for maximizing the use of a potentially bandwidth limited downlink channel. Onboard analysis can also enable rapid reaction to dynamic events, such as flooding, volcanic eruptions or sea ice break-up. Four classifiers were developed to identify cryosphere events using hyperspectral images. These classifiers include a manually constructed classifier, a Support Vector Machine (SVM), a Decision Tree and a classifier derived by searching over combinations of thresholded band ratios. Each of the classifiers was designed to run in the computationally constrained operating environment of the spacecraft. A set of scenes was hand-labeled to provide training and testing data. Performance results on the test data indicate that the SVM and manual classifiers outperformed the Decision Tree and band-ratio classifiers with the SVM yielding slightly better classifications than the manual classifier. The manual and SVM classifiers have been uploaded to the EO-1 spacecraft and have been running onboard the spacecraft for over a year. Results of the onboard analysis are used by the Autonomous Sciencecraft Experiment (ASE) of NASA’s New Millennium Program onboard EO-1 to automatically target the spacecraft to collect follow-on imagery. The software demonstrates the potential for future deep space missions to use onboard decision making to capture short-lived science events.

In this paper, we explore one aspect of knowledge discovery from simulators, the landscape characterization problem, where the aim is to identify regions in the input/parameter/model space that lead to a particular output behavior. Large-scale numerical simulators are in widespread use by scientists and engineers across a range of government agencies, academia, and industry; in many cases, simulators provide the only means to examine processes that are infeasible or impossible to study otherwise. However, the cost of simulation studies can be quite high, both in terms of the time and computational resources required to conduct the trials and the manpower needed to sift through the resulting output. Thus, there is strong motivation to develop automated methods that enable more efficient knowledge extraction. Unlike traditional data mining, knowledge discovery from simulators is not limited to a static, pre-determined dataset; instead, the simulator itself can be used as an oracle to generate new data of our own choosing. We exploit this opportunity by employing active learning and support vector machines (SVMs) to choose which are the most valuable simulation trials to run next. On two realworld scientific simulators, one for asteroid collisions and one for magnetospheric modeling, we demonstrate twofold and sixfold reductions, respectively, in the number of simulator trials required to achieve a particular level of fidelity in landscape characterization as compared with a standard grid-based sampling approach.

SciFlo is a system for Scientific Knowledge Creation on the Grid using a Semantically-Enabled Dataflow Execution Environment. SciFlo leverages Simple Object Access Protocol (SOAP) Web Services and the Grid Computing standards (WS-* & Globus Alliance toolkits), and enables scientists to do multi-instrument Earth Science by assembling reusable Web Services and native executables into a distributed computing flow (operator graph). SciFlo’s XML dataflow documents can be a mixture of concrete operators (fully bound operations) and abstract template operators (late binding via semantic lookup). All data objects and operators can be both simply typed (simple and complex types in XML schema) and semantically typed (linked to OWL ontologies). We will demonstrate a version of the SciFlo workflow engine executing a variety of workflow documents. [Note: This document is a quick description of the proposed demo, not a fully referenced and defended paper.]