As science progresses we witness evolution and revolution in scientific understanding. Coupled intricately with this is an evolution and revolution in scientific techniques and methods – the problem-solving ability of science advances as new methods lead to new understanding, and this in turn leads to new methods…
In the last 10 years science has experienced a step change in problem-solving ability brought about by increasing digitisation, automation and participation. This has in part been achieved through the engagement between scientists and the providers of the advanced information, computational and software techniques that they need. That’s what we call e-Science.
Is it done now? Can we put the “e-” in the history books and just get on with Science?
In a thinktank some months ago about a future strategy for e-Science I argued for “the next level”. e-Science has been successful and, in a properly Darwinian way, is bedding down in our research institutions. But I don’t think we’re done yet – in fact we’re just beginning! What happens next is that we assemble the pieces in order to do entirely new things. That is the next level.
This week I’ve seen a great example of the next-level in action. The music information retrieval community has been developing tools to extract features from digital recordings of music. It’s hard but this multidisciplinary community is flourishing and making great progress. For example, here in Tokyo I’ve just seen (or rather heard…) the vocal extracted from a pop song to generate two files – one with just the vocal, one without. It works incredibly well! Check out the International Society for Music Information Retrieval (ISMIR) and the Music Information Retrieval Evaluation eXchange (MIREX).
And now we’re seeing the research move up a level – having extracted the features, they can be used for the next tier of work; having mastered the computational infrastructure we have it ready to support this. So, for example, you can extract some features and then use autocorrelation to find repeating patterns and figure out the musical stucture – and move from signal processing into musicology. To do this involves remote processing, smart alogorithms and stonking amounts of computation. And it’s a community effort (for content, algorithms and ground truth).
Perhaps the job of work for the science-serving computer scientist is now the assembly rather than the pieces. The music IR community is so ready for assembly- for example, the linked data cloud has substantial music content, thanks to MusicBrainz and the BBC activities. We’ll be exploring this joining-up in the Networked Environment for Music Analysis (NEMA) project.
I wonder what the science history books will say. “1st Generation e-Science 2000-2010 – Accelerated Research”, “2nd Generation e-Science 2010… – stuff that just wouldn’t have happened otherwise”