Dennis DeCoste

Training Invariant Support Vector Machines

Practical experience has shown that in order to obtain the best possible performance, prior knowledge about invariances of a classification problem at hand ought to be incorporated into the training procedure. We describe and review all known methods for doing so in support vector machines, provide experimental results, and discuss their respective merits. One of the significant new results reported in this work is our recent achievement of the lowest reported test error on the well-known MNIST digit recognition benchmark task, with SVM training times that are also significantly faster than previous SVM methods.

Alpha seeding for support vector machines

A key practical obstacle in applying support vector machines to many large-scale data mining tasks is that SVMs generally scale quadratically (or worse) in the number of examples or support vectors.

Rover traverse science for increased mission science return

Rover traverse distances are increasing at a faster rate than downlink capacity is increasing. As this trend continues, the quantity of data that can be returned to Earth per meter traversed is reduced. The capacity of the rover to collect data, however, remains high. Ths circumstance leads to an opportunity to increase mission science return by carefully selecting the data with the highest science interest for downlink. We have developed an onboard science analysis technology for increasing science return from missions. Our technology evaluates the geologic data gather by the rover. This analysis is used to prioritize the data for transmission, so that the data with the highest science value is transmitted to Earth. In addition, the onboard analysis results are used to identify science opportunities. A planning and scheduling component of the system enables the rover to take advantage of the identified science opportunity. Although our techniques are applicable to a wide range of data modalities, our initial emphasis has focused on image analysis, since images consume a large percentage of downlink bandwidth. We have fkther focused our foundational work on rocks. Rocks are among the primary features populating the Martian landscape. Characterization and understanding of rocks on the surface is a-first step leading towards more complex in situ regional geological assessmeats by the rover. IEEEAC paper #1267, Updated November 3,2002 TABLE OF CONTENTS

Distortion-invariant recognition via jittered queries

This paper presents a new approach for achieving distortion-invariant recognition and classification. A test example to be classified is viewed as a query intended to find similar examples in the training set (or to find similar class models that represent a compression of the training set). The key idea is that instead of querying with a single pattern, we construct a more robust query, based on the family of patterns formed by distorting the test example. Although query execution is slower than if the invariances were successfully pre-compiled during training, there are significant advantages in several contexts: (i) providing invariances in memory-based learning, (ii) in model selection, where reducing training time at the expense of test time is a desirable trade-off, and (iii) in enabling robust, ad hoc searches based on a single example. Preliminary tests for memory-based learning on the NIST handwritten digit database with a limited set of shearing and translation distortions produced an error rate of 1.35%.

Learning envelopes for fault detection and state summarization

This paper discusses a data mining approach for overcoming common problems with the traditional red-line limit-checking approach to fault detection and state summarization. It essentially involves learning and adapting parametric functions which provide context-sensitive bounds on historic time-series engineering data. Such bounds are suitable as dynamic plug-in replacements for static red-line values. They enable significantly earlier detection while maintaining low false alarm rates. An example is discussed from onboard tests of this technology during the NASA Deep Space 1 (DS1) mission.