Noise-Resistant Classification: Subsymbolic and Hybrid Architectures for Event Classification in Plasma Physics**The authors wish to thank Steve Chien, Richard Doyle, Usama Fayyad and Nora Mainland for their insightful comments on the ideas presented herein and earlier drafts of this paper.

Two major characteristics of data collected by NASA missions are its staggering amounts and the presence of noise. Scientists require tools which will automate data processing tasks, including data decontamination and classification. Presented here are two approaches to automating these tasks: (1) utilizing neural networks in classification tasks despite significant amounts of noise in the data; and (2) combining neural networks' capability to extract features from data in the presence of noise with inductive learning techniques for assisting plasma physicists in classifying fields and particles events. While these learning techniques are being applied to the data from the magnetometer instrument on the Galileo spacecraft, the general approach is applicable to a wide range of engineering problems involving signal processing and classification in the presence of noise. This paper describes ongoing work at the Jet Propulsion Laboratory in using machine learning techniques to automate these data processing tasks.