Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning

Details Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning

Jul 2010

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2010pasp..122..839z&link_type=abstract

Publications of the Astronomical Society of the Pacific, Volume 122, issue 893, pp.839-852

Mathematics

Logic

Data Analysis And Techniques

Scientific paper

The near-infrared spectra of icy volatiles collected from planetary surfaces can be used to infer surface parameters, which in turn may depend on the recent geologic history. The high dimensionality and complexity of the spectral data, the subtle differences between the spectra, and the highly nonlinear interplay between surface parameters make it often difficult to accurately derive these surface parameters. We use a neural machine, with a Self-Organizing Map (SOM) as its hidden layer, to infer the latent physical parameters, temperature and grain size from near-infrared spectra of crystalline HO ice. The output layer of the SOM-hybrid machine is customarily trained with only the output from the SOM winner. We show that this scheme prevents simultaneous achievement of high prediction accuracies for both parameters. We propose an innovative neural architecture we call Conjoined Twins that allows multiple (k) SOM winners to participate in the training of the output layer and in which the customization of k can be limited automatically to a small range. With this novel machine we achieve scientifically useful accuracies, 83.0±2.7% and 100.0±0.0%, for temperature and grain size, respectively, from simulated noiseless spectra. We also show that the performance of the neural model is robust under various noisy conditions. A primary application of this prediction capability is planned for spectra returned from the Pluto-Charon system by New Horizons.

Affiliated with

Also associated with

No associations

Rating

Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Inference of Surface Parameters from Near-Infrared Spectra of Crystalline HO Ice with Neural Learning will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1230736