Researchers Make Mount Etna Sing

The Interfacer writes "Predicting eruptions will become easier now scientists are using technology to translate the patterns in a volcano's behaviour into sound waves. "The research project, which brings together experts from Europe and Latin America, digitally collects geophysical information on seismic movements before using data sonification to transform them into audible sound waves, which can then be 'scored' as melodies. The resulting 'music' is then analysed for patterns of behaviour and used to identify similarities in eruption dynamics and so predict future activity."

Comments

Grant sucker-uppers?

[MikeWasHere05]

I'm thinking the "convert raw data to music and then extract valuable data from music" step is just in there to ooh and ahh the grant boards. How can that be more efficient than just looking at the raw data?

Re:Grant sucker-uppers?

[poopdeville]

The problem is that most pattern analysis algorithms are computationally expensive. Usually on the order of 2^n computations, unless the researcher is particularly clever and managed to use domain specific knowledge to speed the algorithm up. Reducing your data set by a few orders of magnitude can be the difference between running an algorithm in a day and running it until you're dead.

The up-shot is that instead of making the scientist interpret musical patterns for insights into volcanos (or whatever the researcher is studying), the pattern analysis algorithm will do it for him by correllating patterns to physical phenomena. Indeed, even if it were computationally feasible to perform the calculation on the original data set, data smoothing (say, by lumping subsets into discrete classes as in the article or approximating by statistical analysis) is a good idea to help avoid over-fitting the data. This improves predictive robustness, especially for time-series algorithms.