Learning Bayesian networks from data: An information-theory based approach

This paper provides algorithms that use an information-theoretic analysis to learn Bayesian network structures from data. Based on our three-phase learning framework, we develop efficient algorithms that can effectively learn Bayesian networks, requiring only polynomial numbers of conditional independence (CI) tests in typical cases. We provide precise conditions that specify when these algorithms are guaranteed to be correct as well as empirical evidence (from real world applications and simulation tests) that demonstrates that these systems work efficiently and reliably in practice.

Improved data for solar flare X-ray spectral analysis

The inclusion of collisional rates for He-like Fe and Ca ions is discussed with reference to the analysis of solar flare Fe XXV and Ca XIX line emission, particularly from the Yohkoh Bragg Crystal Spectrometer (BCS). The new data are a slight improvement on calculations presently used in the BCS analysis software in that the discrepancy in the Fe XXV y and z line intensities (observed larger than predicted) is reduced. Values of electron temperature from satellite-to-resonance line ratios are slightly reduced (by up to 1 MK) for a given observed ratio. The new atomic data will be incorporated in the Yohkoh BCS databases. The data should also be of interest for the analysis of high-resolution, non-solar spectra expected from the Constellation-X and Astro-E space missions. A comparison is made of a tokamak S XV spectrum with a synthetic spectrum using atomic data in the existing software and the agreement is found to be good, so validating these data for particularly high-n satellite wavelengths close to the S XV resonance line. An error in a data file used for analyzing BCS Fe XXVI spectra is corrected, so permitting analysis of these spectra.

Error sensitive historical GIS: Identifying areal interpolation errors in time series data

Historical GIS has the potential to re-invigorate our use of statistics from historical censuses and related sources. In particular, areal interpolation can be used to create long-run time-series of spatially detailed data that will enable us to enhance significantly our understanding of geographical change over periods of a century or more. The difficulty with areal interpolation, however, is that the data that it generates are estimates which will inevitably contain some error. This paper describes a technique that allows the automated identification of possible errors at the level of the individual data values.