Blind Astronomers

They use sound instead.

How freaking amazing.


Very cool! Thanks for the link.

This reminds me of methods used in big data and machine learning to find interesting correlations or build effective models in large amounts of data. Often there is benefit to mapping/transforming the data from its native space to a space with a different number of dimensions. Sometimes we reduce dimensionality, sometimes we increase it. Transforming non-sound data into sound is probably an example of this.

When we reduce dimensionality it can help to show associations. For example, lets say we run a clustering algorithm on a data set with 100 dimensions (100 columns in a spreadsheet). It’s very hard to visualize how well this clustering worked in 100 dimensions. If we transform the data to a 2 dimensional space it becomes much easier to see how well the clustering separated various classes etc.

When we increase dimensionality the hope is usually to make the data easier to separate (we want it to be linearly separable). Think of a square in 2-dimensions, with one class (call it A) in upper left corner and lower right and the other class (call it B) in the remaining 2 corners. It’s not possible to draw a straight line that separates these 2 classes. If we project this data to a higher dimensional space, however, we may be able to separate the classes with a flat hyper-plane (multi-dimensional equivalent of a straight line). Maybe when we project to 3 dimensions the A’s take on a z-value of 1, and the B’s take on a z-value of 0 - for example.

I’d be interested in whether there is a benefit to transforming this astronomical data specifically into sound data, or if similar is achieved through more generic forms of increasing/decreasing dimensionality.

1 Like

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.