Last week I attended the music and machine learning (MML) workshop and the subsequent ACM Multimedia (ACM-MM) conference, which were both held in Florence (Italy).
The MML workshop was a well-organized event where one could talk and interact cordially with other members of our scientific community. The scientific content was well-balanced and, overall, the level of the contributions was good. Interestingly, apart from only-symbolic or only-audio-based approaches, there were a couple of papers addressing the fusion of these two aspects: one presented by Carlos Pérez (Pérez et al., Harmonic and instrumental information fusion for musical genre classification) and the other presented by Rudolf Mayer (Mayer et al., Feature selection in a cartesian ensemble of feature subspace classifiers for music categorisation). The fusion aspect is interesting, since I believe that it is an ill-studied aspect within our scientific community.
Apart from general considerations, I would particularly highlight the contribution by Geoffroy Peeters (Peeters, "Copy and scale" method for doing time-localized MIR estimation: application to beat-tracking). Indeed, it seems that the approach Geoffroy is taking is really novel, by the way he looks at the problem, and promising, by the results that are (apparently easily) achieved. The paper is a must for people working in beat detection. Furthermore, it can result in a very inspiring reading for other MIR researchers. There is a forthcoming IEEE-TASLP paper related to the same topic which, if it is as good as the presentation, it promises to be a "killer paper". The MML proceedings will hopefully be on-line soon.
The ACM-MM conference was a big event with more than 600 attendees, the big majority of them coming from Asian countries. The main topics were focused on image and video. However, there was a full session devoted to music ("Novel aids for music retrieval") and some other few scattered music or audio-related papers. Unfortunately, the overall level of these music or audio-related contributions was somehow low. In addition to this, it was sad to witness that the ACM authors working on audio or music have little knowledge of our music information processing community, and that there is little communication between the two. For instance, I was very surprised to hardly see any reference to ICASSP, ISMIR, ICMC, or SMC papers. I also was quite astonished when one presenter had serious difficulties in explaining what a chroma/PCP feature is.
Maybe the most remarkable thing about music-related papers is that one of them was nominated as candidate for the best paper award (Jiajun Bu et al., Music recommendation by unified hypergraph: combining social media information and music content). The approach from Bu et al. seems sound but, since the presentation was not that good, I'll have to read the paper in detail to have a clear opinion about their work.
What I really enjoyed from ACM-MM were the keynote talks, all of them very inspiring. I would specially highlight the one by Duncan Watts, whose presentation pivoted around 4 lines of their more-or-less-recent research: small-world phenomena, social influence and social markets, networks and diffusion, and networked experiments. In particular, related to influence and social markets, he reported results on a "music lab" experiment, where they addressed how the popularity of songs, i.e. number of previous downloads, affects users' ratings of these songs. I will certainly try to read the references he cited (Salganik et al., Experimental study of inequality and unpredictability in an artificial cultural market and Salganik & Watts Leading the herd astray: an experimental study of self-fulfilling prophecies in an artificial cultural market).
Finally, it is worth mentioning the panel "The use of non-conventional means for content media analysis and understanding". This panel was organized on the basis of individual presentations by several group leaders about the "non-conventional" research carried at their labs, plus a short discussion. Some examples of "non-conventional" approaches included image retrieval by brain signals (very impressive), exploitation of eye-gaze information (also quite impressive), research on personality traits, and haptics-related stuff. The ACM-MM proceedings are available on-line here.